Skip Navigation
Skip to contents

ACC : Acute and Critical Care

OPEN ACCESS
SEARCH
Search
Advanced Search
mobile menu button

Articles

Page Path
HOME > Acute Crit Care > Volume 41(1); 2026 > Article
Review Article
Neurology Revolutionizing non-traumatic acute care: a review of the role of artificial intelligence and machine learning in triaging and diagnosis
Acute and Critical Care 2026;41(1):68-86.
DOI: https://doi.org/10.4266/acc.002200
Published online: November 24, 2025

1Department of Internal Medicine, One Brooklyn Health, Brooklyn, NY, USA

2Department of Medicine, Kasturba Medical College, Mangalore, India

3Department of Medicine, Malla Reddy Medical College for Women, Hyderabad, India

4Department of Medicine, Universidad de Ciencias Medicas, San José, Costa Rica

5Department of Emergency Medicine, Green City Hospital, Saharanpur, India

6Department of Radiodiagnosis, Subharti Medical College, Meerut, India

7Department of Medicine, Kettering General Hospital, Kettering, UK

8Department of Medicine, American University of Antigua, Antigua, Barbuda

9Department of Emergency, Prathima Hospital, Hyderabad, India

10Department of Family Medicine, Exceptional Medical Ambulance and Healthcare Services, Dubai, United Arab Emirates

11Department of Medicine, Jinnah Postgraduate Medical Center, Karachi, Pakistan

Corresponding author: Humza Faisal Siddiqui Department of Medicine, Jinnah Postgraduate Medical Center, JPMC, Rafiqui, Sarwar Shaheed Rd, Karachi cantonment, Karachi 75510, Pakistan Tel: +92-336-350-5488 Email: hamsid2024@gmail.com
• Received: July 7, 2025   • Revised: August 31, 2025   • Accepted: September 7, 2025

© 2026 The Korean Society of Critical Care Medicine

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

  • 2,550 Views
  • 75 Download
  • 1 Web of Science
  • 1 Crossref
  • 1 Scopus
prev next
Full Article

Download PDF Download PDF

  • Acute care settings, including emergency medicine and intensive care units, comprise a substantial portion of healthcare and are essential in the prompt management of conditions that can prove fatal. Critical care conditions require timely management that can be delayed by high patient volumes and the need for complex clinical decision making. Artificial intelligence (AI) tools have been created to enhance diagnostic accuracy and optimize workflow to improve patient care. This narrative review discusses the current status of AI in acute care, with a focus on its applications in triaging and diagnosis. AI-enhanced electrocardiogram analysis, identification of myocardial infarction and acute coronary syndrome, and heart failure risk stratification led to better patient-specific management and improved results. AI models successfully determined and aided in the timely management of various acute conditions, including pneumonia, pulmonary embolism, and respiratory failure. The AI algorithms used accurately determined sepsis onset and course, superseding traditionally used clinical tools and leading to early diagnosis and reduced sepsis mortality. These models showed high sensitivity and specificity in diagnosing and triaging neurological conditions, including altered levels of consciousness, seizures, and intracranial hemorrhages. AI that involved advanced machine learning imaging software led to faster and more accurate stroke diagnosis. Diagnostic tools assisted by AI improved the detection and classification of acute pancreatitis, appendicitis, and gastrointestinal bleeding. AI has shown promising results in optimizing management in acute care settings. However, critical issues in data standardization, ethical considerations, and clinical workflow integration need to be addressed to enable clinical implementation.
  • Key Words: appendicitis; artificial intelligence; critical care surgery; myocardial infarction; pulmonary embolism; stroke
Acute care is a vital component of the healthcare system that address episodes of injury or illness that require immediate intervention to prevent death or disability. It encompasses emergency medicine, trauma care, pre-hospital emergency care, acute care surgery, critical care, urgent care, and short-term inpatient stabilization. Neurological, cardiac, and respiratory illnesses; sepsis; and gastrointestinal emergencies are the most common events encountered in acute care settings [1]. Approximately 850,000 emergency department (ED) visits in the United States (US) are affected by sepsis each year [2]. Chest pain remains the second most common reason for adult ED visits in the US, accounting for more than 7 million annual encounters [3]. According to a recent study, more than 795,000 people in the US suffer stroke annually, and every 40 seconds a patient is rushed to an ED due to stroke [4]. In these critical situations, even minor delays or inaccuracies in diagnosis can significantly affect patient outcomes. Accurate and timely diagnosis is essential for providing effective care in acute medical settings. However, acute care settings face multiple challenges, such as managing high patient volumes and rapidly deteriorating patient clinical conditions that require time-sensitive decision-making and prioritization [5].
Artificial intelligence (AI), the simulation of human intelligence by a system or a machine, appears to be a promising solution for the challenges faced in acute care settings. By using data-driven algorithms, AI improves diagnostic accuracy, aids in triaging, reduces human error, and supports quick decision-making in acute care settings. Machine learning (ML) is a branch of AI focused on enabling computers and machines to imitate how humans learn, perform tasks autonomously, and improve their own performance and accuracy through experience and exposure to more data. Deep learning (DL) is a branch of ML that has become increasingly important when using AI for emergency medicine. DL algorithms make it possible to work with data such as medical texts and images. That progress has led to new AI applications for managing patients with urgent needs, including interpreting medical images, automating documentation, and monitoring vital signs in real time [5-7]. Numerous applications, including diagnostic tools, predictive analytics, and personalized medicine, have been created using these principles. AI algorithms excel in accurately analyzing medical images for disease diagnosis, which facilitates the development of personalized treatment plans based on patient data insights. Predictive analytics help identify high-risk patients for timely interventions, and AI-powered tools can optimize workflows, enhancing both efficiency and patient care. Moreover, AI-driven robotics streamline task automation and improve care delivery, particularly in rehabilitation and surgical procedures [8,9]. AI systems can evaluate chest X-rays (CXRs), computed tomography (CT) scans, and magnetic resonance images more quickly, reliably, and accurately than human radiologists, cutting down on turnaround times (TATs). In simulated trials, DL algorithms showed impressive speed, evaluating images up to 150 times faster than radiologists. Additionally, AI facilitates the prompt identification of bad results, which is very helpful in environments with limited resources. ML algorithms are used in cardiology to spot patterns in electrocardiogram (ECG) data that are beyond human perception, such as low ejection fractions or hypoglycemia. ML algorithms, as opposed to rule-based algorithms, can discover patterns directly from data, which gives them great flexibility in applications [10,11].
Figure 1 provides an overview of how AI and ML are used in acute care settings. Overall, the application of AI and ML tools in acute care settings has the potential to improve patient outcomes, enhance diagnostic accuracy, and optimize treatments. However, challenges remain, including limited data for training algorithms, risks of bias, and the need to integrate these tools into clinical workflows. Ethical concerns, such as data privacy and informed consent, also require careful attention [12]. In this study, we review the current literature, explore applications of AI, ML, and DL for diagnosing and triaging in acute care settings, and highlight their potential limitations and challenges.
Institutional Review Board approval and informed consent were not required because this study was a narrative review and did not involve direct data collection from human participants.
A preliminary literature search was performed to find the medical conditions for which most AI and ML tools were built. After that, a more focused literature search was performed on each of those medical conditions. The literature search was performed in PubMed, Scopus, Web of Science, and Google Scholar. The key words used for the literature search in different combinations were acute care, critical care, emergency care, ED, and AI, ML, computer aided design, AND diagnosis, triaging, AND cardiovascular disease, myocardial infarction, heart failure (HF), neurological diseases, stroke, seizure, altered level of consciousness (ALOC), non-traumatic intracranial hemorrhage (ICH), acute respiratory disease, dyspnea, pneumonia, pulmonary embolism, abdominal pain, acute pancreatitis (AP), acute appendicitis (AA), acute gastrointestinal bleeding (GIB), and sepsis. The inclusion criteria were studies published in the English language between 2018 and 2025 that included both male and female patients. Studies including only pediatric patients (younger than 18 years) were excluded.
AI in Cardiovascular Disease Assessment
Cardiovascular diseases remain the leading cause of morbidity and mortality worldwide, and cardiac emergencies require rapid and accurate diagnosis to improve patient outcomes [13]. Cho et al. [14] demonstrated a DL algorithm (DLA) that successfully detected myocardial infarction (MI) in both 12-lead and 6-lead ECGs by using a variational autoencoder to reconstruct precordial ECG from limb ECG. The DLA achieved an area under the receiver operating characteristic curve (AUROC) of 0.902 and 0.901 during internal and external validation, respectively. Lin et al. [15] showed that AI-ECG reduced both door-to-balloon and ECG-to-balloon times, leading to lower in-hospital and 6-month mortality rates. Herman et al. [16] developed an AI model that detected occluded MI with better accuracy than the ST-segment elevation myocardial infarction (STEMI) criteria and ECG experts. It achieved an area under the curve (AUC) of 0.938, with an accuracy of 90.9%, sensitivity of 80.6%, and specificity of 93.7%. They concluded that this AI model could enhance acute coronary syndrome (ACS) triage and help ensure timely revascularization. Doudesis et al. [17] developed the Collaboration for the Diagnosis and Evaluation (CoDE) ACS score, which uses ML, continuous troponin levels, and clinical factors to assess MI probability. Compared with traditional methods, CoDE-ACS improved MI detection accuracy and identified more low-risk patients, i.e., those with a minimal risk of cardiac death within 1 year. Butler et al. [18] developed AI models that used ECG data to predict the 10-year risk of HF and achieved accuracy comparable to or better than current HF risk calculators and traditional ECG methods. Sveric et al. [19] demonstrated that AI-driven automated workflows for left ventricular ejection fraction estimation significantly reduced measurement variability to below 5%, offering greater reliability than traditional methods. This technology enhances the clinical utility of echocardiography by minimizing operator-dependent inconsistencies and providing accurate results across a broad spectrum of patients with varying cardiac conditions (Table 1).
AI in Respiratory Disease Assessment
AI tools showed an accuracy of 93.8%, similar to that of radiologists, in detecting coronavirus disease 2019 (COVID-19) and bacterial pneumonia in CXRs taken in an emergency setting [20]. Predictive models, such as naïve Bayes for CT scans and nearest neighbor for CXRs, accurately identified Pneumocystis jirovecii pneumonia in acquired immunodeficiency syndrome patients in an emergency setting [21]. A convolutional neural network (CNN) assessed CXRs and accurately differentiated COVID-19 pneumonia from community acquired pneumonia [22]. Some physicians found that an AI algorithm could also find subtle consolidations in COVID-19 pneumonia that otherwise went unnoticed [23]. AI models built using CXRs and clinical variables showed better foresight than human experts when predicting the outcomes of COVID-19 patients in the ED [24]. Lin et al. [15] used the blood culture prediction index (BCPI) to create an ML model and tested it by retrospectively analyzing pneumonia patients who presented to an ED. The model performance was analogous to that of CURB-65 in predicting 30-day and in-hospital mortality from among all the risk categories. Subsequently, a Cox regression model that integrated CURB-65 and the BCPI was created and slightly improved the AUC from 0.668 to 0.713. The Cox regression model surpassed CURB-65 in a low-risk group by predicting a substantially lower mortality rate (2.9% vs 7.7%, P<0.001). Thus, conventional clinical tools currently provide the optimum assessment, and AI tools need further development before they can provide superior performance in clinical settings [25].
In one study, a fusion model outperformed an ECG and electronic health record (EHR) model and clinical scoring systems in detecting pulmonary embolism (PE) [26]. Quantitative electrocardiography-right ventricular dysfunction effectively identified right ventricular dysfunction and pulmonary hypertension in patients with PE, outperforming both troponin I and pro-B-type natriuretic peptide [27]. A commercial computer-aided algorithm displayed poor performance in analyzing CT pulmonary angiography, with a low sensitivity of 47% at the lobar level and 50% at the subsegmental lobe level. The tool labeled 92% of the findings as false positives. Thus, in a large ED setting, that algorithm faces several challenges that need to be addressed in future developments [28]. On the other hand, another study reported that a Conformite Europeenne-certified and U.S. Food and Drug Administration (FDA) -approved Al algorithm produced only two false positives, whereas the attending physician produced nine false positives [29]. Integration of an AI tool that analyzed contrast-enhanced CT significantly improved accuracy in detecting incidental PE over the standard procedure, aiding radiologists in triaging patients in an ED [30]. A study that used both internal and external data sets to test the PENet model successfully detected PE precisely, regardless of the slice thickness and manufacturer type [31]. An ML model that analyzed D-dimer and clinical variables showed performance superior to conventional scoring systems when stratifying the risks of patients with suspected PE [32].
In a retrospective study, AI models identified pneumothorax and tension pneumothorax with a high degree of accuracy, so they might be useful for screening in EDs, potentially decreasing the time required to receive reports, such critical X-rays [33]. AI models trained using clinical data and high-resolution bio-signals performed better than conventional methods in diagnosing acute respiratory disease syndrome patients in an intensive care unit (ICU) and stratifying their risk [34,35]. AI has also shown value in predicting the need for mechanical ventilation due to respiratory failure, potentially enabling early intubation in the ED [36]. Deep neural networks successfully and accurately differentiated pulmonary edema from pneumonia in patients older than 65 years [37] (Table 2).
AI in Stroke Assessment
A stroke is a neurological deficit the occurs due to two main causes: brain ischemia as a result of thrombosis, embolism, or systemic hypoperfusion and ICH. Rapid large vessel occlusion (LVO) and Viz LVO are AI imaging software built to identify suspected LVOs in the brain, especially in the distal internal carotid artery and the middle cerebral artery, and thereby assist in the rapid detection and triaging of stroke patients [38,39]. Paz et al. [38] used rapid LVO to analyze the CT scans of 151 patients and reported a sensitivity of 63.6% and specificity of 85.8%. The software demonstrated a mean notification time of 32.5 minutes. AI and ML tools exhibited substantial accuracy in triaging and diagnosing stroke in a systematic review of 30 studies performed by Shlobin et al. [39]. Elijovich et al. [40] found that implementation of the Viz AI reduced the time needed to notify management teams, which allowed patients to receive earlier thrombectomies than with standard protocols. Gunda et al. [41] analyzed e-Stroke Suite software to streamline decision-making in a primary stroke center and reported that it reduced the time to receive thrombectomy and thrombolysis therapy, improving patient outcomes.
The Brainomix electronic-computed tomography angiography (CTA) is AI-driven decision-enhancement software that helps to assess collateral circulation and thus predict patient outcomes after thrombectomy. The Brainomix e-CTA is FDA cleared and CE marked and uses efficacious algorithms and large data analyses for LVO identification. It has a sensitivity of 83.8% (95% CI, 77.3%–88.7%), and its specificity is 95.7% (95% CI, 91.0%–98.0%) [42]. Avicenna CINA is an AI application used to detect both ischemic and hemorrhagic strokes from CT angiography images. It also recognizes LVOs, ICHs, and other cerebrovascular accident, including cerebrovascular diseases, to help in the clinical management of emergencies. Avicenna CINA has been found to be highly reliable, with a sensitivity of 98.1% (95% CI, 94.5%–99.3%) and a specificity of 98.2% (95% CI, 95.5%–99.3%), making it one of the most precise tools available for stroke diagnosis [43,44]. A systematic review of all studies showed that AI software generally enhanced diagnostic accuracy and improved management. However, concerns about cost-effectiveness persist [45]. In particular, lesions observed on diffusion weighted imaging (DWI) are an indicator of a recent stroke. A multivariate model designed to evaluate DWI showed an AUC of 0.8 and positive predictive value of 0.90 in identifying LVOs [46] (Table 3).
AI in Neurological Disease Assessment
Neurological emergencies account for approximately 20% of emergency room patients. For clinicians without significant neurologic expertise, epilepsy is difficult to diagnose in certain situations. McInnis et al. [47] demonstrated that AI, ML, and DL significantly improve the speed and precision of emergency neurological interventions by automating seizure monitoring in electroencephalogram (EEG) data, especially in resource-scarce locations. AI-EEG has a higher accuracy rate than clinical experts, with a sensitivity of 84% and specificity of 72% based on clinical factors alone, and a sensitivity of 92% and specificity of 83% when the data are weighted. Tveit et al. [48] evaluated SCORE-AI, which interprets clinical EEGs with a validity equal to human experts, with an accuracy of 88.3% (79.2–94.9%), and thereby reduces diagnostic errors and improves interrater agreement. Keikhosrokiani et al. [49] proposed AI and ML models that significantly increased the speed of seizure detection and diagnosis. The accuracy of a random forest (RF) model reached 96.52% with an AUROC of 99.03%, and a CNN model obtained a validation accuracy of 97.65%. Therefore, these algorithms are valid for the rapid detection and treatment of seizures and offer considerable efficiency improvements in emergency care. SPaRCNet is a massive training dataset that correctly classifies electrographic seizures and abnormalities. One study emphasized that AI datasets such as SPaRCNet, along with other sophisticated ML classifiers, markedly enhance the precision of seizure detection by mitigating interrater variability and bolstering predictive functions [50]. AI and ML technologies have not yet shown performance metrics robust enough to change current clinical practices. However, future innovations in cloud computing will likely enable the quantitative assessment of epileptiform abnormalities, network dynamics, and the spatial localization of epileptogenic zones, thus improving patient outcomes [51].
Seyam et al. [52] reported an AI-doc medical application with a diagnostic accuracy of 93.0% and a high negative predictive value of 97.8% for ICH overall, although the sensitivity in identifying subarachnoid hemorrhage was insufficient. Matsoukas et al. [53] recently demonstrated that the aggregate sensitivity, specificity, and accuracy rates for AI tools that identify ICH and cerebral microbleeds exceeded 92%. Heit et al. [54] validated the RAPID ICH tool and reported very high accuracy in detecting and quantifying the volume of ICH, including a sensitivity is 95.6% and specificity of 95.3%. Wang et al. [55] used advanced neural networks with interpretability features such as Grad-CAM to improve the accuracy and transparency of AI diagnostics in ICH detection. Wismüller and Stockmaster [56] proposed the AI-PROBE framework that successfully reduced important TATs and identified ICH without compromising on precision. Savage et al. [57] tested diagnostic accuracy and reported TATs with AI assistance that were on par with radiologists, which might suggest that the benefit is limited in practice, however impressive it appears to be in theory. Agarwal et al. [58] pointed to the need for better validation and understanding of the clinical effects of AI, underlining a pooled sensitivity and specificity of 0.90 for ICH detection in CT imaging. Davis et al. [59] reported that AI-doc produced significant reductions in report TAT and improved the length of stay for patients, particularly those without ICH, indicating that ML has the potential to improve clinical management. Voter et al. [60] reported a sensitivity of 92.3%, specificity of 97.7%, positive predictive value of 81.3%, and negative predictive value of 99.2% for AI-doc. However, its performance in cases with a prior history of neurosurgery for ICH was weak. Overall, the accuracy of AI tools for the detection of ICH is substantially good. AI applications improve ICH management efficiency by shortening the time needed to scan reviews and prioritizing severe cases. Ginat et al. [61] evaluated the ability of the AI-doc tool to identify critically acute ICH cases based on head CT scans and reported that AI implementation decreased scan view delays for flagged outpatient and inpatient cases.
Hsu et al. [62] indicated that ML models, such as RF and logistic regression, improved the triage process by forecasting patient outcomes, which enhanced operational efficiency in evaluating an ALOC. Their study indicates that ML models that incorporate natural language processing significantly enhanced the speed and accuracy of predicting patient dispositions, which improved emergency care efficiency and timely management of ALOC. DL models using a CNN architecture (Word2vec) with three different types of word embeddings were able to automate the identification of altered mental status in the clinical notes of an ED, with AUROC values as high as 98.5% and accuracy of 94.5% [63]. Another study demonstrated that the validated performance of a model based on the MIMIC III dataset produced high accuracy values: mean absolute error, 0.269; mean squared error, 0.625; and coefficient of determination, 0.964. Therefore, the ML models significantly enhanced the predictive and intuitive explanation of consciousness levels, compared with traditional methods [64] (Table 4).
AI in Gastrointestinal Disease Assessment
AP is a sudden and severe inflammation of the pancreas that is usually triggered by gallstones or excessive alcohol intake and requires immediate medical attention. To establish proper treatment and monitoring, AP patients need rapid diagnosis and risk classification. Zhou et al.’s systematic review of 24 studies that evaluated 47 ML models demonstrated that the models [65] were effective in diagnosing AP and predicting mortality, recurrence, and timing for surgical intervention. Another recent study showed that AI-based tools improved pancreatic measurement and segmentation, which enhanced the diagnosis efficacy and workflow of radiologists in AP and AP with pancreatic ductal adenocarcinoma cases [66]. Kui et al. [67] conducted an early achievable severity index study using an AI tool that identified patients with a high risk of severe AP with an accuracy of 89.1% and improved with experience. Podda et al. [68] reported that combining human knowledge and AI helped to efficiently manage acute biliary pancreatitis in EDs and ICUs. Similarly, AI models in another study predicted the severity index and need for ICU in 1334 patients with an accuracy of 88% and 98%, respectively [69].
AA is one of the most common causes of abdominal pain in the ED. Park at al. [70] analyzed abdominal CT scans and reported that their CNN-based algorithm diagnosed appendicitis with more than 90% accuracy among patients presenting to an ED with severe abdominal pain. A systemic analysis of 29 studies reflected that artificial neural networks (ANNs) displayed superiority over conventional methods in diagnosing and prognosing AA [71]. Akbulut et al. [72] reported that AI models were able to differentiate perforated AA from non-perforated AA with an accuracy of more than 90%. Roshanaei et al. [73] used AI and ML tools to analyze patients who presented with acute abdominal pain in the ED. Their Gaussian naïve Bayes model showed dominant performance, with an accuracy of 95.03%, sensitivity of 87.18%, and specificity of 97.54% in diagnosing AA. A gradient boosting algorithm was the second most accurate, with an accuracy of 94.4%. The RF model and support vector classifier exhibited slightly inferior performance, with accuracies of 92.55% and 91.93%, respectively, in diagnosing AA. Ghareeb et al. [74] performed a multicenter retrospective cohort study to evaluate the performance of an AI platform in diagnosing AA. The platform exhibited a sensitivity of 92.2%, specificity of 97.2%, and negative predictive value of 98.7%.
The risk stratification and decision-making process to diagnose acute GIB in the ED can be streamlined by implementing AI models [75]. Shung et al. [76] conducted a systematic analysis of 14 studies that tested 30 ML models for evaluating overt GIB cases. Overall, the ML tools proved to be more efficient than clinical tools in assessing mortality from upper GIB, and the ANN showed superiority over other models. Even though small bowel endoscopy is minimally invasive and has a high clinical result probability, evaluation of the images is time consuming and requires human effort, which has a scope of error. A CNN-based model diagnosed the risk of potential bleeding by analyzing capsule endoscopy findings with a sensitivity of 88% and specificity of 99% [77]. A pooled study reflected that CNN-based computer-aided diagnosis showed excellent execution in identifying gastrointestinal ulcerations and hemorrhages when evaluating wireless capsule endoscopy images [78] (Table 5).
AI in Sepsis Assessment
Sepsis is a life-threatening condition in which the body has an exaggerated immune response to an infection, and therefore early detection and intervention are key in preventing mortality. Approximately one in five patients who present to the ED with a suspicion of sepsis do not show any signs of organ dysfunction, although their clinical condition begins to deteriorate within 48 hours of admission. Barton et al. created an algorithm to estimate the onset of sepsis 48 hours in advance using just vital signs, and it had an AUC of 0.83 [79]. Delahanty et al. [80] used an ML model to create a screening tool called the Risk of Sepsis (RoS) score, which proved to be more sensitive and precise than conventional tools. There was an increase in the AUROC of the RoS score from 0.93 in the first hour to 0.97 after 24 hours. Using a different data set from an ICU, Mao et al. [81] achieved an AUROC of 0.9238 for the InSight algorithm, which diagnosed sepsis 4 hours before onset. Additionally, 4 hours before onset, the InSight algorithm estimated septic shock with an AUROC of 0.96. Kaji et al. [82] presented an AI tool that was able to predict same day and following day sepsis onset and the need for vancomycin administration among patients admitted to an ICU. Taneja et al. [83] used vital signs and biomarkers such as procalcitonin to estimate sepsis onset and reported an AUROC of 0.81. Henry et al. [84] created a real-time warning score to estimate the onset of sepsis a median of 28.2 hours before onset and reported an AUROC of 0.83. Schinkel et al. [85] documented eleven different models that predicted sepsis with AUROCs in the range of 0.696–0.952, mostly in ED and ICU populations. Those models outperformed existing tools for diagnosing sepsis, specifically the Sequential Organ Failure Assessment (SOFA), Systemic Inflammatory Response Syndrome criteria (SIRS) and Modified Early Warning score (MEWS). The InSight algorithm achieved an AUROC of 0.880, whereas the other models were remarkably lower: SOFA (0.725), SIRS (0.609) and MEWS (0.803) for the same population [85]. Yuan et al. [86] performed a prospective open cohort study that compared an AI algorithm for sepsis diagnosis with the SOFA score-based diagnostic method. Their study used 1,588 incidents (444 sepsis and 1,144 non-sepsis) in patients with a mean age of 67.6 years and APACHE II scores of 13.8. The AI algorithm showed an accuracy of 82% ± 1%, sensitivity of 65% ± 5%, specificity of 88% ± 2%, precision of 67%±3%, and F1 of 0.66±0.02. The AI algorithm was superior to the SOFA score-based method, with AUCs of 0.89 and 0.596, respectively. Another systematic analysis of nine studies revealed that ML models were effective for the early diagnosis of sepsis [87]. Yue et al. [88] demonstrated that ML models are a reliable tool for predicting acute kidney injury in patients with sepsis. Zhang et al. [89] used ML models that evaluated clinical markers to predict mortality among patients with sepsis. Lactate, which is usually ignored in clinical tools, was highlighted as a potential indicator of sepsis-related mortality. She et al. [90] used ML algorithms to evaluate the association between various metabolites and sepsis and potentially help with diagnosis and guide management (Table 6).
Limitations and Ethical Challenges of AI
Randomized evidence is beginning to emerge. For example, the AI–Powered Rapid Identification of ST-Elevation Myocardial Infarction via Electrocardiogram (ARISE) trial confirmed that AI-ECG interpretation shortened reperfusion times and reduced mortality from STEMI, and the Viz AI and e-Stroke Suite accelerated stroke triage and thrombectomy workflows [15,40]. AI consistently exceeded the predictive accuracy of widely used scoring tools, such as SOFA, CURB-65, and traditional troponin-based algorithms, suggesting that it provides a meaningful advantage in triage and risk stratification [85-88]. Although AI reduces TATss and optimizes workflow, early commercial systems generated high false-positive burdens. Newer FDA-cleared models demonstrate substantially improved specificity, though careful monitoring is still required to prevent alarm fatigue. AI systems should be positioned as adjuncts, not replacements for human expertise. Maintaining clinician oversight ensures that trust and shared decision-making remain central to patient care. Although workflow improvements are evident, robust economic analyses remain sparse. Future multicenter randomized controlled trials must incorporate cost-utility endpoints to establish whether AI adoption is financially sustainable. Potential improvements include federated learning, harmonization of data standards (HL7 FHIR, DICOM), and validation using international multicenter datasets to reduce bias and enhance generalizability [91-93].
AI has the potential to improve efficiency, patient outcomes, and access to care. However, several challenges remain. Medical data are significantly more complex than data in other domains and encompass diverse formats, such as text, images, numerical values, classifications, and diagnostic information. Healthcare data are often fragmented and lack standardization, and incomplete datasets are a common challenge. Additionally, they include genetic sequencing data, details about procedures performed, and the results of various investigations. This complexity creates a paradox: despite having extensive information about a patient, the comprehensive data needed to fully leverage machine-driven insights are lacking. The complexity of medical decision-making, which is driven by the need to account for numerous interacting factors, adds a layer of difficulty not present in other fields. Data used across different operating systems are often challenging to integrate, making it difficult to achieve standardized results. Different guidelines are fed into various operating systems, which makes the goal of achieving standardization in AI-assisted healthcare delivery a significant challenge [91,92]. Additionally, the performance of AI tools can vary when they are applied to different populations or clinical settings, which can lead to inconsistent results. For example, many studies reported in this review demonstrated high diagnostic accuracy, but they were not externally validated across institutions, limiting the generalizability of those results to real-world emergency clinical practice. The quality and quantity of data play a crucial role in the planning and effectiveness of AI tools to ensure optimal outcomes. One significant issue is algorithmic bias: many AI systems fail to account for ethnicity, sex, and other sociodemographic characteristics, potentially leading to inequities in healthcare delivery. Operator dependency and interobserver variability can also affect the performance of AI tools [93-95]. Although AI has reduced those issues in some areas, such as echocardiography, they still pose a problem in other domains in which the quality of the imaging or subtle findings can influence diagnostic performance. Another challenge faced by healthcare workers who want to use AI algorithms is the lack of high-quality evidence to support the results. Randomized controlled trials and meta-analyses are often insufficient when validating AI applications due to inadequate sample sizes and limited study periods. Although numerous studies present high AUC and accuracy metrics, many of them used retrospective single-center datasets and lack prospective validation and multi-institutional testing, which calls their reproducibility and generalizability into question. To ensure better validation, studies should include diverse locations worldwide and encompass different races, ethnicities, and parts of society. Situational awareness is another area in which AI systems are lacking, particularly in acute care environments. Algorithmic outputs can oversimplify complex clinical decisions, a phenomenon called computational reductionism. Such reduction can lead to a lack of consideration of individual patient nuances and the broader clinical context, potentially resulting in suboptimal or even harmful recommendations. Consequently, human judgment remains mandatory in clinical decision-making, even after input from an AI model. The integration of AI into healthcare requires significant changes to existing systems. This includes acceptance by clinicians, modifications to instruments to ensure compatibility with AI systems, and careful consideration of cost-benefit analyses. Early AI technologies face challenges such as interoperability and integration with existing EHR systems. They often have limited clinical utility and are associated with the black box phenomenon, in which the lack of transparency in AI decision-making makes it difficult to interpret or trust the outputs. These systems often rely on retrospective testing methods, which might not adequately translate in prospective clinical trials. Only a small fraction, around 2%, of AI models progress beyond the prototyping phase, highlighting the challenges institutions face in translating AI advances into real-world clinical settings. The lack of required infrastructure in most emergency settings also hinders the clinical implementation of these tools. Data privacy and compliance with regulations are critical concerns. Ensuring that data are collected, processed, and stored securely and in accordance with regulatory standards is paramount to maintaining trust and efficacy. Recent studies indicate that 51% of healthcare service users express concerns about data privacy. Lack of collaboration leads to disparities in decision-making between AI systems and healthcare workers, as well as misalignment with societal expectations. Lastly, public acceptance of AI in healthcare is influenced by these challenges, which underscores the importance of transparency, explainability, and inclusive engagement to ensure that technological advances align with ethical and societal values [91-97].
Acute care faces critical challenges in patient management, such as accurate diagnosis, risk assessment, and prompt management. AI has shown significant potential to help overcome these obstacles and improve patient outcomes. AI has demonstrated competency in augmenting diagnostic capabilities for myocardial infarctions and stroke, prognostication of the onset of sepsis, and improvement in respiratory condition protocols. However, it is also important to consider its limitations, including the generalizability, external validation, algorithmic bias, and ethical concerns of many AI models. Future directions include prospective multicenter trials, integration with EHR systems, and the development of transparent and explainable models that support, rather than replace, clinical decision-making procedures. Because certain challenges persist, continuous research to improve ethical data flow and collaboration among clinicians, AI specialists, and policy makers will be essential to fully capitalize on AI’s potential and ensure that it translates into an effective tool in acute care setting.
• Artificial intelligence (AI) and machine learning (ML) increase accuracy in diagnosing medical conditions in acute care and emergency settings, thereby enhancing patient outcomes.
• AI and ML optimize workflow and reduce the time needed to triage and diagnose patients in acute care and emergency settings, thus helping to provide timely patient care.

CONFLICT OF INTEREST

No potential conflict of interest relevant to this article was reported.

FUNDING

None.

ACKNOWLEDGMENTS

None.

AUTHOR CONTRIBUTIONS

Conceptualization: OD, RMS, PB, MGC, AUH, SH, MI, AIC, SMY, HAWB, HFS. Methodology: OD, RMS, PB, MGC, AUH, SH, MI, AIC, SMY, HAWB, HFS. Validation: OD, RMS, PB, MGC, AUH, SH, MI, AIC, SMY, HAWB, HFS. Investigation: OD, RMS, PB, MGC, AUH, SH, MI, AIC, SMY, HAWB, HFS. Data curation: OD, RMS, PB, MGC, AUH, SH, MI, AIC, SMY, HAWB, HFS. Visualization: OD, RMS, PB, MGC, AUH, SH, MI, AIC, SMY, HAWB, HFS. Supervision: HFS. Project administration: OD, RMS, PB, MGC, AUH, SH, MI, AIC, SMY, HAWB, HFS. Writing–original draft: OD, RMS, PB, MGC, AUH, SH, MI, AIC, SMY, HAWB, HFS. Writing–review & editing: OD, RMS, PB, MGC, AUH, SH, MI, AIC, SMY, HAWB, HFS. All authors read and agreed to the published version of the manuscript.

Figure 1.
Overview of role of artificial intelligence in acute and critical care. CT: computed tomography; MRI: magnetic resonance imaging; ARDS: acute respiratory distress syndrome. This figure was created using Biorender.com by Humza Faisal Siddiqui and Omofolarin Debellotte.
acc-002200f1.jpg
acc-002200f2.jpg
Table 1.
Overview of studies using AI to diagnose cardiovascular diseases
Study AI/ML tool used Disease/symptom Data used AUC (95%CI) Outcome
Butler et al. (2023) [18] AI-based ECG models 10-Year heart failure risk ECG data, patient history and examination findings, BMI and vital signs ECG-chars: 0.73 (0.70–0.77) Comparable accuracy than current HF risk calculators. Improved long-term risk prediction
ARIC HF-risk calculator: 0.76 (0.72–0.80)
FH-HF risk calculator: 0.74 (0.70–0.78)
CPH: 0.78 (0.75–0.80)
ECG-AI: 0.77 (0.74–0.79)
ECG-AI-cox: 0.84 (0.81–0.87)
Cho et al. (2020) [14] DLA with VAE MI 12-Lead and 6-lead ECG Internal validation: 0.88 ML algorithms improved MI detection. Six-lead DLA with VAE outperformed non-VAE
External validation: 0.854
Doudesis et al. (2023) [17] Collaboration for the Diagnosis and Evaluation of ACS score using ML Acute MI Continuous troponin levels, patient history, clinical factors. 0.953 (0.947–0.958) Improved MI probability detection; identified more low-risk patients with minimal 1-year cardiac death risk
Herman et al. (2023) [16] AI model Occluded MI ECG data 0.938 (0.924–0.951) Enhanced ACS triage with high accuracy
Lin et al. (2024) [15] AI-ECG ST-elevation myocardial infarction ECG data - Reduced door-to-balloon and ECG-to-balloon times. Decrease in-hospital and 6-month mortality
Sveric et al. (2024) [19] AI-driven automated workflows HF and left ventricular ejection fraction estimation Echocardiography images Auto-ECHO: 0.93 Reduced measurement variability (<5%). Improved reliability over traditional methods
MBS-ECHO: 0.92

AI: artificial intelligence; ML: machine learning; AUC: area under the curve; ECG: electrocardiogram; BMI: body mass index; ARIC: atherosclerosis risk in communities; HF: heart failure; FH: familial hypercholesterolemia; CPH: cox proportional hazards; DLA: deep-learning algorithm; VAE: variational autoencoder; MI: myocardial infarction; ACS: acute coronary syndrome; ECHO: echocardiogram; MBS: medicare benefits schedule echocardiography.

Table 2.
Overview of studies using AI to diagnose respiratory diseases
Study AI/ML tool used Disease/symptom Data used Accuracy measure Outcome
Ippolito et al. (2023) [20] Multi-tasking mask R CNN Pneumonia CXR Accuracy: 93.8% Remarkable accuracy of AI tool found in differentiating between COVID-19 pneumonia, typical bacterial pneumonia, and healthy subjects
Chagas et al. (2024) [21] Predictive models using Boruta algorithm AIDS- associated PCP LDH, O2 saturation AUC: >0.8 Predictive models differentiated AIDS PCP
CRP, RR (>24 bpm) dry cough, HIV viral load, CD4 cell count and CXRs
Salvatore et al. (2021) [22] ResNet- 50 architecture and trained and cross validated COVID-19 pneumonia CXRs AUC: 0.98 AI tools showed potential in correctly differentiating COVID-19 from community acquired pneumonia.
Carlile et al. (2020) [23] CNN COVID-19 pneumonia CXRS with heat maps - More than 80% physicians found the tool aided in optimizing the workflow and clinical decisions.
Kwon et al. (2020) [24] DenseNet-121 architecture pretrained on ImageNet COVID-19 pneumonia Chest radiographs and clinical variables AUC: Model improved prediction regarding mortality and intubation requirement.
 0.88 for intubation
 0.82 for mortality
Lin et al. (2024) [25] Blood culture prediction index and CURB-65 integrated novel Pneumonia CBC, differential leukocyte counts, BUN, age, sex, RR, blood pressure, GCS AUC: 0.713 for cox regression model Integrated models had better predictive performance for low risk patients.
COX regression model
Somani et al. (2021) [26] Fusion model- ML model Pulmonary embolism Clinical data and ECGs AUC: 0.84 AUC for fusion model Fusion models performed better than the other models in improving screening of acute PE.
Choi et al. (2024) [27] Smartphone application ECG BUDDY Pulmonary embolism ECGs AUC: 0.895 Right ventricular dysfunction identified rapidly on ECG as compared to markers like Troponin I and ProBNP.
QCG-RVDys
Müller-Peltzer et al. (2021) [28] Computer aided algorithm integrated into Siemens Healthineers software, VB20A Pulmonary embolism CT pulmonary angiography Sensitivity: 47% at lobar and 50% at subsegmental level Computer-aided algorithms lead to higher false positives can be used to assist radiologists but needs radiological examination for confirmation.
Langius-Wiffen et al. (2023) [29] AI algorithm Pulmonary embolism CT pulmonary angiography Sensitivity: 96.8% Al algorithm showed higher accuracy in detecting pulmonary embolism versus radiologist.
Specificity: 99.9%
Savage et al. (2024) [30] AI triage system, the BriefCase for iPE from the vendor Aidoc Pulmonary embolism Contrast enhanced CT scans Sensitivity: 96.2 AI assistance showed increased sensitivity for detecting pulmonary embolism. However, no significant change observed in decreasing time.
Specificity: 99.9
Huang et al. (2020) [31] Deep learning 77- layer 3D convolutional neural network (PENet) Pulmonary embolism CT scan AUC: Successful automated application for diagnosis of pulmonary embolism with the AI tool.
0.84 for internal validation
0.85 for external validation
Villacorta et al. (2022) [32] Generalized learning logistic regression using elastic net Pulmonary embolism O2 saturation, history of deep venous thrombosis or pulmonary embolism, immobilization or surgery and D dimer AUC: 0.89 D dimer maintained a low false negative rate
Hillis et al. (2022) [33] AI model Pneumothorax CXR AUC: Al model detected pneumothorax and tension pneumothorax accurately.
 0.97 for pneumothorax
 0.98 for tension pneumothorax
Han et al. (2025) [34] Convolutional Recurrent Neural Network and sequence modelling (recurrent neural network) ARDS ECG, HR, RR, SpO2, and non-invasive systolic and diastolic blood pressures, body temperature, AVPU score, age, sex, arterial blood gas analysis AUC: Positive prediction scores and clinical outcomes were established with the AI model.
 0.84 for internal validation
 0.73 for external validation
Rashid et al. (2022) [35] Multiple AI models ARDS Variable AUC: 0.8–1 AI models reduced cost and improved outcomes amongst ARDS patients.
Viderman et al. (2024) [36] Feed forward neural networks, CNN and ANN ARDS Variable - AI models shown to perform well in predicting respiratory failure.
Liong-Rung et al. (2022) [37] GoogleNet Inception V4 architecture Dyspnea CXRs Accuracy: 83.2% High negative predictive value of excluding pulmonary edema & high positive predictive value in diagnosing pneumonia.

AI: artificial intelligence; ML: machine learning; CNN: convolutional neural network; CXR: chest X-ray; COVID-19: coronavirus disease 2019; AIDS: acquired immunodeficiency syndrome; PCP: pneumocystis jirovecii pneumonia; LDH: lactate dehydrogenase; CRP: C-reactive protein; RR: respiratory rate; HIV: human immunodeficiency virus; CD: cluster of differentiation; AUC: area under the curve; CBC: complete blood count; BUN: blood urea nitrogen; GCS: Glasgow coma scale; ECG: electrocardiogram; PE: pulmonary embolism; ProBNP: pro-B-type natriuretic peptide; CT: computed tomography; iPE: incidental pulmonary embolism; 3D: three-dimensional; ARDS: acute respiratory distress syndrome; HR: heart rate; AVPU: alert, verbal, pain, unresponsive; ANN: artificial neural networks.

Table 3.
Overview of studies using AI to diagnose stroke
Study AI/ML model Disease Data used Accuracy measure Outcome
Westwood et al. (2024) [45] Viz LVO, Rapid LVO, Brainomix, Avicenna Stroke Head CTA Sensitivity: 95.4% The study showed that AI software enhanced accuracy in stroke detection.
Specificity: 79.4%
Elijovich et al. (2022) [40] Viz AI Stroke Head CTA - The AI software reduced time to notify management team and patients received earlier thrombectomies.
Gunda et al. (2022) [41] e-Stroke Suite Stroke Non-contrast Head CT and CTA - The AI system helped in decision making process. Reduced time and improved rate of receiving reperfusion treatment strategies.
Shlobin et al. (2022) [39] CNN, RF and SVM Stroke CTA Sensitivity: 63.6% AI tools accurately triaged and diagnosed patients with large vessel occlusion.
Specificity: 85.8%
Paz et al. (2021) [38] Rapid LVO Stroke Head CT scan - The study showed that AI tool had low sensitivity and moderate specificity.
Amukotuwa et al. (2019) [44] Avicenna CINA Stroke Head CT angiography Sensitivity: 95% Avicenna CINA proved to be one of the most accurate tools in stroke detection.
Specificity: 79%
Schmitt et al. (2022) [42] Brainomix e-CTA Stroke Non-contrast enhanced head CT Sensitivity: 0.91 The AI-based algorithm dependably determined if acute intra-cranial hemorrhage were present amongst participants and calculated intra-parenchymal hemorrhage volumes accurately
Specificity: 0.89
Wouters et al. (2018) [46] Automated multi-variate imaging model Stroke Diffusion weighted and perfusion weighted imaging AUC: 0.80 The model accurately identified patients of large vessel occlusion within 6-hour window.

AI: artificial intelligence; ML: machine learning; LVO: large vessel occlusion; CTA: computed tomography angiography; CT: computed tomography; CNN: convolutional neural network; RF: random forest; SVM: support vector machine; AUC: area under the curve.

Table 4.
Overview of studies using AI to diagnose neurological diseases
Study AI/ML/DL tool used Disease/symptom Data used Accuracy measure Outcome
McInnis et al. (2023) [47] AI-EEG Epilepsy EEG AUC: 0.86 Higher diagnostic accuracy than clinical experts
Tveit et al. (2023) [48] SCORE-AI Epilepsy EEG Accuracy: 88.3% Reduces diagnostic errors and improves interrater agreement
Keikhosrokiani et al. (2024) [49] AI augmented pathway for digital care pathway for epilepsy (DCPE) Epilepsy Scalp EEG Accuracy: 97.65% for CNN Increase the speed of seizure detection and diagnosis
Jahan et al. (2023) [51] Cloud computing Epilepsy EEG - Assessment of epileptiform abnormalities and localization of epileptogenic zones
Seyam et al. (2022) [52] AI-doc Hemorrhage Head CT Accuracy: 93% High diagnostic and predictive value for ICH
Matsoukas et al. (2022) [53] Logistic regression, support machine vector, random forest, gradient boosting, deep CNN Hemorrhage NCCT and MRI Accuracy: 93.46% for ICH Detected ICH and microbleeds with high accuracy
Heit et al. (2021) [54] RAPID-ICH Hemorrhage NCCT scans Sensitivity: 0.95 High accuracy in detecting and quantifying the volume of ICH
Specificity: 0.95
Wang et al. (2021) [55] GRAD-CAM Hemorrhage CT scans - Improve the accuracy and transparency in ICH detection
Wismüller et al. (2020) [56] AI-PROBE Hemorrhage CT scans Accuracy: 96.4% TAT reduction for cases of ICH
Savage et al. (2024) [57] AI-doc (commercial AI triage system) Hemorrhage NCCT Accuracy: 99.2% Diagnostic accuracy and reported TAT with AI assistance was on par with radiologists
Agarwal et al. (2023) [58] Convolutional neural network, AI-doc, Avicenna.ai Hemorrhage CT and MRI Sensitivity: 0.90 Reflects that most abnormality detection AI studies were not adequately validated in representative clinical cohorts
Specificity: 0.90
Davis et al. (2022) [59] AI-doc Hemorrhage CT scans - Reported significant reductions in RTAT along with improved LOS
Voter et al. (2021) [60] (AI) decision support systems, Aidoc Hemorrhage CT scans Sensitivity: 92.3% Diagnostic performance in cases with a prior history of neurosurgery ICH is weaker
Specificity: 97.7%
Ginat et al. (2021) [61] AI-doc Hemorrhage CT - Shortened the time needed to scan reviews and prioritize severe cases
Hsu et al. (2023) [62] Random forest and logistic regression ALOC due to hyperglycemia Electronic medical records AUC: 0.79 for all-cause mortality Improve the triage process by forecasting patient outcomes
Obeid et al. (2019) [63] Word2vec Altered mental status Patient notes Accuracy: 94.5% To automate the identification of altered mental status
El-Rashidy et al. (2023) [64] MIMIC III ALOC EEG data and vital signs Mean absolute error: 0.269 Enhance the predictive and intuitive explanation of consciousness levels

AI: artificial intelligence; ML: machine learning; DL: deep learning; EEG: electroencephalogram; AUC: area under the curve; CNN: convolutional neural network; CT: computed tomography; ICH: intracranial hemorrhage; NCCT: non-contrast computed tomography; MRI: magnetic resonance imaging; TAT: turnaround time; LOS: length of stay; RTAT: report turnaround time; ALOC: altered level of consciousness.

Table 5.
Overview of studies using AI to diagnose gastro-intestinal diseases
Study AI/ML tool Disease Data used Accuracy measure Outcome
Zhou et al. (2022) [65] 47 ML models (ANN, LR, SVM, XGBoost, and RF) AP Clinical, laboratory and imaging parameters - ML models showed potential in diagnosing AP and predicting mortality and recurrence.
Pan et al. (2024) [66] MSAnet (CNN) AP and AP&PDAC Non-contrast and enhanced abdominal CT AUC: 0.99 AI tools enhanced diagnostic accuracy.
Kui et al. (2022) [67] EASY-APP (XGBoost) AP RR, body temperature, abdominal muscle reflex, sex, age, and blood glucose levels Accuracy: 89.1% The AI tool predicted risk of severe AP within few hours of hospital admission with high accuracy.
Podda et al. (2024) [68] ChatGPT Acute biliary pancreatitis Clinical and laboratory values - The AI tool helped optimize the management of patients in ED and ICU.
İnce et al. (2023) [69] Gradient boost ML algorithm AP Demographic, clinical and laboratory parameters Accuracy: The ML tool proficiently delineated the severity, need of ICU and survival among patients.
 98.25% for ICU need 92.77% for survival
Park et al. (2020) [70] CNN AA Abdominal CT Accuracy: more than 90% The ML tool diagnosed AA among patient presenting in ED with severe abdominal pain with high accuracy.
Issaiy et al. (2023) [71] ANN and LR AA Clinical, laboratory and imaging parameters AUC: 0.985 AI tools diagnosed AA and predicted post-surgical risk of sepsis and ICU requirement efficiently.
Akbulut et al. (2023) [72] CatBoost model Perforated and non-perforated AA CBC, bilirubin, CRP, age and other laboratory values. Accuracy: 92% The AI model distinguished perforated and non-perforated AA with remarkable accuracy.
Roshanaei et al. (2024) [73] GNB model, RF model, GBA and SVM AA Patient characteristics, laboratory parameters and cause of pain Accuracy: AI tools showed promising results in accurately diagnosing AA in the emergency setting. Gaussian naïve bayes model displayed most superior results.
 95.03% for GNB
 92.55% for RF
 94.41% for GBA
 91.93% for SVM
Ghareeb et al. (2024) [74] AI platform AA Patient characteristics AUC: 0.97 The AI platform showed to be an effective diagnostic tool in detecting AA.
Shung at al. (2019) [76] 30 ML models GIB Clinical parameters AUC: ML models showed superiority over clinical tools in predicting mortality in upper GIB. ANN proved to be the most effective model.
 0.93 for ANN
 0.81 for other ML models
Saraiva et al. (2021) [77] CNN GIB Capsule endoscopy images Accuracy: 99% Model proved to be effective in identifying small bowel lesions and predicting risk of bleeding.
Mohan et al. (2021) [78] CNN-based CAD GIB or gastrointestinal hemorrhage WCE Accuracy: 95.4% CNN models showed high accuracy, NPV and PPV in diagnosing GIB or hemorrhage by proficiently interpreting WCE images.

AI: artificial intelligence; ML: machine learning; ANN: artificial neural network; LR: logistic regression; SVM: support vector machine; RF: random forest; AP: acute pancreatitis; CNN: convoluted neural network; AP&PDAC: acute pancreatitis with pancreatic ductal carcinoma; CT: computed tomography; AUC: area under the curve; RR: respiratory rate; ED: emergency department; ICU: intensive care unit; AA: acute appendicitis; CBC: complete blood count; CRP: C-reactive protein; GNB: Gaussian naïve bayes; GBA: gradient boost algorithm; GIB: gastrointestinal bleeding; ANN: artificial neural network; CAD: computer aided design; WCE: wireless capsule endoscopy NPV: negative predictive value.

Table 6.
Overview of studies using AI to diagnose sepsis
Study AI/ML tool used Disease Data used Accuracy measure Outcome
Delahanty et al. (2019) [80] Risk of Sepsis score Sepsis Rhee clinical surveillance criteria AUC: 0.93–0.97 ML model proved to be more effective than conventional clinical screening tools.
Kaji et al. (2019) [82] Long short-term memory recurrent Neural Network Sepsis Clinical parameters AUC: The model effectively predicted daily sepsis occurrence among patients in admitted in ICU.
 0.87 for sepsis
 0.83 for antibiotic administration
Barton et al. (2019) [79] Supervised gradient -boosted tree model Sepsis Vital signs AUC: 0.88 for sepsis onset The model accurately predicted sepsis 48 hours before onset.
Mao et al. (2018) [81] InSight Severe sepsis and septic shock Vital signs Accuracy: ML model predicted severe sepsis and septic shock 4 hours before onset with high accuracy.
 0.92 for sepsis
 0.87 for severe sepsis
Henry et al. (2022) [84] TRWES ML-based warning system Sepsis Clinical parameters - The alert system model decreased time to initiate antibiotic treatment and improved patient outcomes.
Yuan et al. (2020) [86] Diagnostic algorithm Sepsis Clinical data Accuracy: 82% The AL algorithm can improve patient outcomes by diagnosing sepsis among ICU patients accurately as compared to SOFA.
XG Boost
Yan et al. (2022) [87] ML models Sepsis Clinical data notes from providers, demographic, vital signs, laboratory data and medications - The models proved effective in early identification of sepsis.
Yue et al. (2022) [88] ANN, XGBoost, LR and SVM AKI in sepsis Various clinical and laboratory parameters Accuracy: 0.82 for XGboost The ML models proved to be proficient in predicting AKI among patients with sepsis.
Zhang et al. (2023) [89] Multiple ML models Mortality in sepsis Various clinical and laboratory parameters Sensitivity: 0.71 The ML models showed superiority in predicting mortality as compared to conventional clinical scoring tools. ML highlighted lactate to be an effective indictor, usually missed by clinical tools.
Specificity: 0.68
She et al. (2023) [90] SVM and RF Sepsis Non-targeted liquid chromatography-high-resolution mass spectrometry metabolomics - ML model delineated the association between various metabolites and sepsis that can potentially aid in diagnosis and management.

AI: artificial intelligence; ML: machine learning; AUC: area under the curve; ICU: intensive care unit; SOFA: Sequential Organ Failure Assessment; ANN: artificial neural network; LR: logistic regression; SVM: support vector machine; AKI: acute kidney injury; RF: random forest.

  • 1. Hirshon JM, Risko N, Calvello EJ, Stewart de Ramirez S, Narayan M, Theodosis C, et al. Health systems and services: the role of acute care. Bull World Health Organ 2013;91:386-8.ArticlePubMedPMC
  • 2. Wang HE, Jones AR, Donnelly JP. Revised national estimates of emergency department visits for sepsis in the United States. Crit Care Med 2017;45:1443-9.ArticlePubMedPMC
  • 3. Cairns C, Kang K. National hospital ambulatory medical care survey: 2019 emergency department summary tables [Internet]. National Center for Health Statistics 2022 [cited 2025 Sep 15]. Available from: www.cdc.gov/nchs/products/databriefs/db434.htm
  • 4. Tsao CW, Aday AW, Almarzooq ZI, Anderson CA, Arora P, Avery CL, et al. Heart disease and stroke statistics-2023 update: a report from the American Heart Association. Circulation 2023;147:e93-621.ArticlePubMedPMC
  • 5. Boonstra A, Laven M. Influence of artificial intelligence on the work design of emergency department clinicians a systematic literature review. BMC Health Serv Res 2022;22:669.ArticlePubMedPMCPDF
  • 6. Xu Y, Liu X, Cao X, Huang C, Liu E, Qian S, et al. Artificial intelligence: a powerful paradigm for scientific research. Innovation (Camb) 2021;2:100179.ArticlePubMedPMC
  • 7. Piliuk K, Tomforde S. Artificial intelligence in emergency medicine. A systematic literature review. Int J Med Inform 2023;180:105274.ArticlePubMed
  • 8. Johnson KB, Wei WQ, Weeraratne D, Frisse ME, Misulis K, Rhee K, et al. Precision medicine, AI, and the future of personalized health care. Clin Transl Sci 2021;14:86-93.ArticlePubMedPDF
  • 9. Olawade DB, David-Olawade AC, Wada OZ, Asaolu AJ, Adereni T, Ling J, et al. AI in healthcare delivery: Prospects and pitfalls. J Med Surg Public Health 2024;3:100108.
  • 10. Aggarwal R, Sounderajah V, Martin G, Ting DS, Karthikesalingam A, King D, et al. Diagnostic accuracy of deep learning in medical imaging: a systematic review and meta-analysis. NPJ Digit Med 2021;4:65.ArticlePubMedPMCPDF
  • 11. Quer G, Arnaout R, Henne M, Arnaout R. Machine learning and the future of cardiovascular care: JACC state-of-the-art review. J Am Coll Cardiol 2021;77:300-13.ArticlePubMedPMC
  • 12. Sreedharan JK, Saleh F, Alqahtani A, Albalawi IA, Gopalakrishnan GK, Alahmed HA, et al. Applications of artificial intelligence in emergency and critical care diagnostics: a systematic review and meta-analysis. Front Artif Intell 2024;7:1422551.ArticlePubMedPMC
  • 13. Sun X, Yin Y, Yang Q, Huo T. Artificial intelligence in cardiovascular diseases: diagnostic and therapeutic perspectives. Eur J Med Res 2023;28:242.ArticlePubMedPMCPDF
  • 14. Cho Y, Kwon JM, Kim KH, Medina-Inojosa JR, Jeon KH, Cho S, et al. Artificial intelligence algorithm for detecting myocardial infarction using six-lead electrocardiography. Sci Rep 2020;10:20495.ArticlePubMedPMCPDF
  • 15. Lin C, Liu WT, Chang CH, Lee CC, Hsing SC, Fang WH, et al. AI-Powered Rapid Identification of ST-Elevation Myocardial Infarction via Electrocardiogram (ARISE) — a pragmatic randomized controlled trial. NEJM AI 2024;1:1-12.
  • 16. Herman R, Meyers HP, Smith SW, Bertolone DT, Leone A, Bermpeis K, et al. International evaluation of an artificial intelligence-powered electrocardiogram model detecting acute coronary occlusion myocardial infarction. Eur Heart J Digit Health 2024;5:123-33.ArticlePubMedPDF
  • 17. Doudesis D, Lee KK, Boeddinghaus J, Bularga A, Ferry AV, Tuck C, et al. Machine learning for diagnosis of myocardial infarction using cardiac troponin concentrations. Nat Med 2023;29:1201-10.ArticlePubMedPMC
  • 18. Butler L, Karabayir I, Kitzman DW, Alonso A, Tison GH, Chen LY, et al. A generalizable electrocardiogram-based artificial intelligence model for 10-year heart failure risk prediction. Cardiovasc Digit Health J 2023;4:183-90.ArticlePubMedPMC
  • 19. Sveric KM, Ulbrich S, Dindane Z, Winkler A, Botan R, Mierke J, et al. Improved assessment of left ventricular ejection fraction using artificial intelligence in echocardiography: a comparative analysis with cardiac magnetic resonance imaging. Int J Cardiol 2024;394:131383.ArticlePubMed
  • 20. Ippolito D, Maino C, Gandola D, Franco PN, Miron R, Barbu V, et al. Artificial intelligence applied to chest X-ray: a reliable tool to assess the differential diagnosis of lung pneumonia in the emergency department. Diseases 2023;11:171.ArticlePubMedPMC
  • 21. Chagas OJ, Gonçalves FA, Nagatomo PP, Buccheri R, Pereira-Chioccola VL, Del Negro GM, et al. Predictive models-assisted diagnosis of AIDS-associated Pneumocystis jirovecii pneumonia in the emergency room, based on clinical, laboratory, and radiological data. Sci Rep 2024;14:11247.ArticlePubMedPMCPDF
  • 22. Salvatore C, Interlenghi M, Monti CB, Ippolito D, Capra D, Cozzi A, et al. Artificial intelligence applied to chest X-ray for differential diagnosis of COVID-19 pneumonia. Diagnostics (Basel) 2021;11:530.ArticlePubMedPMC
  • 23. Carlile M, Hurt B, Hsiao A, Hogarth M, Longhurst CA, Dameff C, et al. Deployment of artificial intelligence for radiographic diagnosis of COVID-19 pneumonia in the emergency department. J Am Coll Emerg Physicians Open 2020;1:1459-64.ArticlePubMedPMCPDF
  • 24. Kwon YJ, Toussie D, Finkelstein M, Cedillo MA, Maron SZ, Manna S, et al. Combining initial radiographs and clinical variables improves deep learning prognostication in patients with COVID-19 from the emergency department. Radiol Artif Intell 2021;3:e200098. ArticlePubMed
  • 25. Lin YT, Lin KM, Wu KH, Lien F. Enhancing pneumonia prognosis in the emergency department: a novel machine learning approach using complete blood count and differential leukocyte count combined with CURB-65 score. BMC Med Inform Decis Mak 2024;24:118.ArticlePubMedPMCPDF
  • 26. Somani SS, Honarvar H, Narula S, Landi I, Lee S, Khachatoorian Y, et al. Development of a machine learning model using electrocardiogram signals to improve acute pulmonary embolism screening. Eur Heart J Digit Health 2022;3:56-66.ArticlePubMedPDF
  • 27. Choi YJ, Park MJ, Cho Y, Kim J, Lee E, Son D, et al. Screening for RV dysfunction using smartphone ECG analysis app: validation study with acute pulmonary embolism patients. J Clin Med 2024;13:4792.ArticlePubMedPMC
  • 28. Müller-Peltzer K, Kretzschmar L, Negrão de Figueiredo G, Crispin A, Stahl R, Bamberg F, et al. Present limitations of Artificial Intelligence in the emergency setting - performance study of a commercial, computer-aided detection algorithm for pulmonary embolism. Rofo 2021;193:1436-44.ArticlePubMed
  • 29. Langius-Wiffen E, de Jong PA, Hoesein FA, Dekker L, van den Hoven AF, Nijholt IM, et al. Retrospective batch analysis to evaluate the diagnostic accuracy of a clinically deployed AI algorithm for the detection of acute pulmonary embolism on CTPA. Insights Imaging 2023;14:102.ArticlePubMedPMCPDF
  • 30. Savage CH, Elkassem AA, Hamki O, Sturdivant A, Benson D, Grumley S, et al. Prospective evaluation of artificial intelligence triage of incidental pulmonary emboli on contrast-enhanced CT examinations of the chest or abdomen. AJR Am J Roentgenol 2024;223:e2431067. ArticlePubMed
  • 31. Huang SC, Kothari T, Banerjee I, Chute C, Ball RL, Borus N, et al. Penet-a scalable deep-learning model for automated diagnosis of pulmonary embolism using volumetric CT imaging. NPJ Digit Med 2020;3:61.ArticlePubMedPMCPDF
  • 32. Villacorta H, Pickering JW, Horiuchi Y, Olim M, Coyne C, Maisel AS, et al. Machine learning with D-dimer in the risk stratification for pulmonary embolism: a derivation and internal validation study. Eur Heart J Acute Cardiovasc Care 2022;11:13-9.ArticlePubMedPDF
  • 33. Hillis JM, Bizzo BC, Mercaldo S, Chin JK, Newbury-Chaet I, Digumarthy SR, et al. Evaluation of an artificial intelligence model for detection of pneumothorax and tension pneumothorax in chest radiographs. JAMA Netw Open 2022;5:e2247172. ArticlePubMedPMC
  • 34. Han C, Jung YJ, Park JE, Chung WY, Yoon D. Artificial intelligence-based early prediction of acute respiratory failure in the emergency department using biosignal and clinical data. Yonsei Med J 2025;66:121-30.ArticlePubMedPDF
  • 35. Rashid M, Ramakrishnan M, Chandran VP, Nandish S, Nair S, Shanbhag V, et al. Artificial intelligence in acute respiratory distress syndrome: a systematic review. Artif Intell Med 2022;131:102361.ArticlePubMed
  • 36. Viderman D, Ayazbay A, Kalzhan B, Bayakhmetova S, Tungushpayev M, Abdildin Y, et al. Artificial intelligence in the management of patients with respiratory failure requiring mechanical ventilation: a scoping review. J Clin Med 2024;13:7535.ArticlePubMedPMC
  • 37. Liong-Rung L, Hung-Wen C, Ming-Yuan H, Shu-Tien H, Ming-Feng T, Chia-Yu C, et al. Using artificial intelligence to establish chest X-ray image recognition model to assist crucial diagnosis in elder patients with dyspnea. Front Med (Lausanne) 2022;9:893208.ArticlePubMedPMC
  • 38. Paz D, Yagoda D, Wein T. Single site performance of AI software for stroke detection and triage. medRxiv [Preprint]. 2021 [cited 2025 Nov 10]. Available from: https://doi.org/10.1101/2021.04.02.21253083.Article
  • 39. Shlobin NA, Baig AA, Waqas M, Patel TR, Dossani RH, Wilson M, et al. Artificial intelligence for large-vessel occlusion stroke: a systematic review. World Neurosurg 2022;159:207-20.e1.ArticlePubMed
  • 40. Elijovich L, Dornbos Iii D, Nickele C, Alexandrov A, Inoa-Acosta V, Arthur AS, et al. Automated emergent large vessel occlusion detection by artificial intelligence improves stroke workflow in a hub and spoke stroke system of care. J Neurointerv Surg 2022;14:704-8.ArticlePubMed
  • 41. Gunda B, Neuhaus A, Sipos I, Stang R, Böjti PP, Takács T, et al. Improved stroke care in a primary stroke centre using AI-decision support. Cerebrovasc Dis Extra 2022;12:28-32.ArticlePubMedPMCPDF
  • 42. Schmitt N, Mokli Y, Weyland CS, Gerry S, Herweh C, Ringleb PA, et al. Automated detection and segmentation of intracranial hemorrhage suspect hyperdensities in non-contrast-enhanced CT scans of acute stroke patients. Eur Radiol 2022;32:2246-54.ArticlePubMedPDF
  • 43. Amukotuwa SA, Straka M, Dehkharghani S, Bammer R. Fast automatic detection of large vessel occlusions on CT angiography. Stroke 2019;50:3431-8.ArticlePubMedPMC
  • 44. Amukotuwa SA, Straka M, Smith H, Chandra RV, Dehkharghani S, Fischbein NJ, et al. Automated detection of intracranial large vessel occlusions on computed tomography angiography: a single center experience. Stroke 2019;50:2790-8.ArticlePubMed
  • 45. Westwood M, Ramaekers B, Grimm S, Armstrong N, Wijnen B, Ahmadu C, et al. Software with artificial intelligence-derived algorithms for analysing CT brain scans in people with a suspected acute stroke: a systematic review and cost-effectiveness analysis. Health Technol Assess 2024;28:1-204.ArticlePDF
  • 46. Wouters A, Dupont P, Christensen S, Norrving B, Laage R, Thomalla G, et al. Multimodal magnetic resonance imaging to identify stroke onset within 6 h in patients with large vessel occlusions. Eur Stroke J 2018;3:185-92.ArticlePubMedPMCPDF
  • 47. McInnis RP, Ayub MA, Jing J, Halford JJ, Mateen FJ, Brandon Westover M, et al. Epilepsy diagnosis using a clinical decision tool and artificially intelligent electroencephalography. Epilepsy Behav 2023;141:109135.ArticlePubMedPMC
  • 48. Tveit J, Aurlien H, Plis S, Calhoun VD, Tatum WO, Schomer DL, et al. Automated interpretation of clinical electroencephalograms using artificial intelligence. JAMA Neurol 2023;80:805-12.ArticlePubMedPMC
  • 49. Keikhosrokiani P, Isomursu M, Uusimaa J, Kortelainen J. A sustainable artificial-intelligence-augmented digital care pathway for epilepsy: automating seizure tracking based on electroencephalogram data using artificial intelligence. Digit Health 2024;10:20552076241287356.ArticlePubMedPMCPDF
  • 50. Kerr WT, McFarlane KN, Figueiredo Pucci G. The present and future of seizure detection, prediction, and forecasting with machine learning, including the future impact on clinical trials. Front Neurol 2024;15:1425490.ArticlePubMedPMC
  • 51. Jahan S, Nowsheen F, Antik MM, Rahman MS, Kaiser MS, Hosen AS, et al. AI-based epileptic seizure detection and prediction in internet of healthcare things: a systematic review. IEEE 2023;11:30690-725.Article
  • 52. Seyam M, Weikert T, Sauter A, Brehm A, Psychogios MN, Blackham KA, et al. Utilization of artificial intelligence-based intracranial hemorrhage detection on emergent noncontrast CT images in clinical workflow. Radiol Artif Intell 2022;4:e210168. ArticlePubMedPMC
  • 53. Matsoukas S, Scaggiante J, Schuldt BR, Smith CJ, Chennareddy S, Kalagara R, et al. Accuracy of artificial intelligence for the detection of intracranial hemorrhage and chronic cerebral microbleeds: a systematic review and pooled analysis. Radiol Med 2022;127:1106-23.ArticlePubMedPDF
  • 54. Heit JJ, Coelho H, Lima FO, Granja M, Aghaebrahim A, Hanel R, et al. Automated cerebral hemorrhage detection using RAPID. AJNR Am J Neuroradiol 2021;42:273-8.ArticlePubMedPMC
  • 55. Wang Z, Wu L, Ji X. An interpretable deep learning system for automatic intracranial hemorrhage diagnosis with CT image. Proc 2021 Int Conf Bioinform Intell Comput 2021;338-35.Article
  • 56. Wismüller A, Stockmaster L. A prospective randomized clinical trial for measuring radiology study reporting time on AI-based detection of intracranial hemorrhage in emergent care head CT. Inmedical 2020 Biomed Appl Mol Struct Funct 2020;11317:144-50.
  • 57. Savage CH, Tanwar M, Elkassem AA, Sturdivant A, Hamki O, Sotoudeh H, et al. Prospective evaluation of artificial intelligence triage of intracranial hemorrhage on noncontrast head CT examinations. AJR Am J Roentgenol 2024;223:e2431639. ArticlePubMed
  • 58. Agarwal S, Wood D, Grzeda M, Suresh C, Din M, Cole J, et al. Systematic review of artificial intelligence for abnormality detection in high-volume neuroimaging and subgroup meta-analysis for intracranial hemorrhage detection. Clin Neuroradiol 2023;33:943-56.ArticlePubMedPMCPDF
  • 59. Davis MA, Rao B, Cedeno PA, Saha A, Zohrabian VM. Machine learning and improved quality metrics in acute intracranial hemorrhage by noncontrast computed tomography. Curr Probl Diagn Radiol 2022;51:556-61.ArticlePubMed
  • 60. Voter AF, Meram E, Garrett JW, Yu JJ. Diagnostic accuracy and failure mode analysis of a deep learning algorithm for the detection of intracranial hemorrhage. J Am Coll Radiol 2021;18:1143-52.ArticlePubMedPMC
  • 61. Ginat D. Implementation of machine learning software on the radiology worklist decreases scan view delay for the detection of intracranial hemorrhage on CT. Brain Sci 2021;11:832.ArticlePubMedPMC
  • 62. Hsu CC, Kao Y, Hsu CC, Chen CJ, Hsu SL, Liu TL, et al. Using artificial intelligence to predict adverse outcomes in emergency department patients with hyperglycemic crises in real time. BMC Endocr Disord 2023;23:234.ArticlePubMedPMCPDF
  • 63. Obeid JS, Weeda ER, Matuskowitz AJ, Gagnon K, Crawford T, Carr CM, et al. Automated detection of altered mental status in emergency department clinical notes: a deep learning approach. BMC Med Inform Decis Mak 2019;19:164.ArticlePubMedPMCPDF
  • 64. El-Rashidy N, Sedik A, Siam AI, Ali ZH. An efficient edge/cloud medical system for rapid detection of level of consciousness in emergency medicine based on explainable machine learning models. Neural Comput Appl 2023;35:10695-716.ArticlePubMedPMCPDF
  • 65. Zhou Y, Ge YT, Shi XL, Wu KY, Chen WW, Ding YB, et al. Machine learning predictive models for acute pancreatitis: a systematic review. Int J Med Inform 2022;157:104641.ArticlePubMed
  • 66. Pan X, Jiao K, Li X, Feng L, Tian Y, Wu L, et al. Artificial intelligence-based tools with automated segmentation and measurement on CT images to assist accurate and fast diagnosis in acute pancreatitis. Br J Radiol 2024;97:1268-77.ArticlePubMedPMCPDF
  • 67. Kui B, Pintér J, Molontay R, Nagy M, Farkas N, Gede N, et al. EASY-APP: An AI model and application for early and easy prediction of severity in acute pancreatitis. Clin Transl Med 2022;12:e842. ArticlePubMedPMC
  • 68. Podda M, Di Martino M, Ielpo B, Catena F, Coccolini F, Pata F, et al. The 2023 MANCTRA acute biliary pancreatitis care bundle: a joint effort between human knowledge and artificial intelligence (ChatGPT) to optimize the care of patients with acute biliary pancreatitis in Western countries. Ann Surg 2024;279:203-12.ArticlePubMed
  • 69. İnce AT, Silahtaroğlu G, Seven G, Koçhan K, Yıldız K, Şentürk H, et al. Early prediction of the severe course, survival, and ICU requirements in acute pancreatitis by artificial intelligence. Pancreatology 2023;23:176-86.ArticlePubMed
  • 70. Park JJ, Kim KA, Nam Y, Choi MH, Choi SY, Rhie J, et al. Convolutional-neural-network-based diagnosis of appendicitis via CT scans in patients with acute abdominal pain presenting in the emergency department. Sci Rep 2020;10:9556.ArticlePubMedPMCPDF
  • 71. Issaiy M, Zarei D, Saghazadeh A. Artificial intelligence and acute appendicitis: a systematic review of diagnostic and prognostic models. World J Emerg Surg 2023;18:59.ArticlePubMedPMCPDF
  • 72. Akbulut S, Yagin FH, Cicek IB, Koc C, Colak C, Yilmaz S, et al. Prediction of perforated and nonperforated acute appendicitis using machine learning-based explainable artificial intelligence. Diagnostics (Basel) 2023;13:1173.ArticlePubMedPMC
  • 73. Roshanaei G, Salimi R, Mahjub H, Faradmal J, Yamini A, Tarokhian A, et al. Accurate diagnosis of acute appendicitis in the emergency department: an artificial intelligence-based approach. Intern Emerg Med 2024;19:2347-57.ArticlePubMedPDF
  • 74. Ghareeb WM, Draz E, Chen X, Zhang J, Tu P, Madbouly K, et al. Multicenter validation of an artificial intelligence (AI)-based platform for the diagnosis of acute appendicitis. Surgery 2024;176:569-76.ArticlePubMed
  • 75. Shung DL. Advancing care for acute gastrointestinal bleeding using artificial intelligence. J Gastroenterol Hepatol 2021;36:273-8.ArticlePubMedPMCPDF
  • 76. Shung D, Simonov M, Gentry M, Au B, Laine L. Machine learning to predict outcomes in patients with acute gastrointestinal bleeding: a systematic review. Dig Dis Sci 2019;64:2078-87.ArticlePubMedPDF
  • 77. Mascarenhas Saraiva MJ, Afonso J, Ribeiro T, Ferreira J, Cardoso H, Andrade AP, et al. Deep learning and capsule endoscopy: automatic identification and differentiation of small bowel lesions with distinct haemorrhagic potential using a convolutional neural network. BMJ Open Gastroenterol 2021;8:e000753. ArticlePubMedPMC
  • 78. Mohan BP, Khan SR, Kassab LL, Ponnada S, Chandan S, Ali T, et al. High pooled performance of convolutional neural networks in computer-aided diagnosis of GI ulcers and/or hemorrhage on wireless capsule endoscopy images: a systematic review and meta-analysis. Gastrointest Endosc 2021;93:356-64.e4.ArticlePubMed
  • 79. Barton C, Chettipally U, Zhou Y, Jiang Z, Lynn-Palevsky A, Le S, et al. Evaluation of a machine learning algorithm for up to 48-hour advance prediction of sepsis using six vital signs. Comput Biol Med 2019;109:79-84.ArticlePubMedPMC
  • 80. Delahanty RJ, Alvarez J, Flynn LM, Sherwin RL, Jones SS. Development and evaluation of a machine learning model for the early identification of patients at risk for sepsis. Ann Emerg Med 2019;73:334-44.ArticlePubMed
  • 81. Mao Q, Jay M, Hoffman JL, Calvert J, Barton C, Shimabukuro D, et al. Multicentre validation of a sepsis prediction algorithm using only vital sign data in the emergency department, general ward and ICU. BMJ Open 2018;8:e017833. ArticlePubMedPMC
  • 82. Kaji DA, Zech JR, Kim JS, Cho SK, Dangayach NS, Costa AB, et al. An attention based deep learning model of clinical events in the intensive care unit. PLoS One 2019;14:e0211057. ArticlePubMedPMC
  • 83. Taneja I, Reddy B, Damhorst G, Dave Zhao S, Hassan U, Price Z, et al. Combining biomarkers with EMR data to identify patients in different phases of sepsis. Sci Rep 2017;7:10800.ArticlePubMedPMCPDF
  • 84. Henry KE, Adams R, Parent C, Soleimani H, Sridharan A, Johnson L, et al. Factors driving provider adoption of the TREWS machine learning-based early warning system and its effects on sepsis treatment timing. Nat Med 2022;28:1447-54.ArticlePubMedPDF
  • 85. Schinkel M, Paranjape K, Nannan Panday RS, Skyttberg N, Nanayakkara PW. Clinical applications of artificial intelligence in sepsis: a narrative review. Comput Biol Med 2019;115:103488.ArticlePubMed
  • 86. Yuan KC, Tsai LW, Lee KH, Cheng YW, Hsu SC, Lo YS, et al. The development an artificial intelligence algorithm for early sepsis diagnosis in the intensive care unit. Int J Med Inform 2020;141:104176.ArticlePubMed
  • 87. Yan MY, Gustad LT, Nytrø Ø. Sepsis prediction, early detection, and identification using clinical text for machine learning: a systematic review. J Am Med Inform Assoc 2022;29:559-75.ArticlePubMedPDF
  • 88. Yue S, Li S, Huang X, Liu J, Hou X, Zhao Y, et al. Machine learning for the prediction of acute kidney injury in patients with sepsis. J Transl Med 2022;20:215.ArticlePubMedPMCPDF
  • 89. Zhang Y, Xu W, Yang P, Zhang A. Machine learning for the prediction of sepsis-related death: a systematic review and meta-analysis. BMC Med Inform Decis Mak 2023;23:283.ArticlePubMedPMCPDF
  • 90. She H, Du Y, Du Y, Tan L, Yang S, Luo X, et al. Metabolomics and machine learning approaches for diagnostic and prognostic biomarkers screening in sepsis. BMC Anesthesiol 2023;23:367.ArticlePubMedPMCPDF
  • 91. van de Sande D, Chung EF, Oosterhoff J, van Bommel J, Gommers D, van Genderen ME, et al. To warrant clinical adoption AI models require a multi-faceted implementation evaluation. NPJ Digit Med 2024;7:58.ArticlePubMedPMC
  • 92. Pinsky MR, Bedoya A, Bihorac A, Celi L, Churpek M, Economou-Zavlanos NJ, et al. Use of artificial intelligence in critical care: opportunities and obstacles. Crit Care 2024;28:113.ArticlePubMedPMCPDF
  • 93. Thompson CL, Morgan HM. Ethical barriers to artificial intelligence in the National Health Service, United Kingdom of Great Britain and Northern Ireland. Bull World Health Organ 2020;98:293-5.ArticlePubMedPMC
  • 94. Collins BX, Bélisle-Pipon JC, Evans BJ, Ferryman K, Jiang X, Nebeker C, et al. Addressing ethical issues in healthcare artificial intelligence using a lifecycle-informed process. JAMIA Open 2024;7:ooae108.ArticlePubMedPMCPDF
  • 95. Elendu C, Amaechi DC, Elendu TC, Jingwa KA, Okoye OK, John Okah M, et al. Ethical implications of AI and robotics in healthcare: a review. Medicine (Baltimore) 2023;102:e36671. ArticlePubMedPMC
  • 96. Solaiman B. Legal and ethical considerations of artificial intelligence for residents in post-acute and long-term care. J Am Med Dir Assoc 2024;25:105105.ArticlePubMed
  • 97. Farrokhi M, Fallahian AH, Rahmani E, Aghajan A, Alipour M, Jafari Khouzani P, et al. Current applications, challenges, and future directions of artificial intelligence in emergency medicine: a narrative review. Arch Acad Emerg Med 2025;13:e45. Article

Figure & Data

References

    Citations

    Citations to this article as recorded by  
    • Insights into Acute Pancreatitis: Pathogenesis, Diagnosis, and Management
      Silvia Carrara, Federico Cassano, Maria Terrin, Marco Spadaccini
      Journal of Clinical Medicine.2026; 15(8): 2819.     CrossRef

    • PubReader PubReader
    • ePub LinkePub Link
    • Cite
      CITE
      export Copy
      Close
      Download Citation
      Download a citation file in RIS format that can be imported by all major citation management software, including EndNote, ProCite, RefWorks, and Reference Manager.
      Format:
      • RIS — For EndNote, ProCite, RefWorks, and most other reference management software
      • BibTeX — For JabRef, BibDesk, and other BibTeX-specific software
      Include:
      • Citation for the content below
      Revolutionizing non-traumatic acute care: a review of the role of artificial intelligence and machine learning in triaging and diagnosis
      Acute Crit Care. 2026;41(1):68-86.   Published online November 24, 2025
      DOI: https://doi.org/10.4266/acc.002200
      Close
    • XML DownloadXML Download
    Figure
    • 0
    • 1
    Revolutionizing non-traumatic acute care: a review of the role of artificial intelligence and machine learning in triaging and diagnosis
    Image Image
    Figure 1. Overview of role of artificial intelligence in acute and critical care. CT: computed tomography; MRI: magnetic resonance imaging; ARDS: acute respiratory distress syndrome. This figure was created using Biorender.com by Humza Faisal Siddiqui and Omofolarin Debellotte.
    Graphical abstract

    Figure 1.

    Graphical abstract

    Revolutionizing non-traumatic acute care: a review of the role of artificial intelligence and machine learning in triaging and diagnosis
    Study AI/ML tool used Disease/symptom Data used AUC (95%CI) Outcome
    Butler et al. (2023) [18] AI-based ECG models 10-Year heart failure risk ECG data, patient history and examination findings, BMI and vital signs ECG-chars: 0.73 (0.70–0.77) Comparable accuracy than current HF risk calculators. Improved long-term risk prediction
    ARIC HF-risk calculator: 0.76 (0.72–0.80)
    FH-HF risk calculator: 0.74 (0.70–0.78)
    CPH: 0.78 (0.75–0.80)
    ECG-AI: 0.77 (0.74–0.79)
    ECG-AI-cox: 0.84 (0.81–0.87)
    Cho et al. (2020) [14] DLA with VAE MI 12-Lead and 6-lead ECG Internal validation: 0.88 ML algorithms improved MI detection. Six-lead DLA with VAE outperformed non-VAE
    External validation: 0.854
    Doudesis et al. (2023) [17] Collaboration for the Diagnosis and Evaluation of ACS score using ML Acute MI Continuous troponin levels, patient history, clinical factors. 0.953 (0.947–0.958) Improved MI probability detection; identified more low-risk patients with minimal 1-year cardiac death risk
    Herman et al. (2023) [16] AI model Occluded MI ECG data 0.938 (0.924–0.951) Enhanced ACS triage with high accuracy
    Lin et al. (2024) [15] AI-ECG ST-elevation myocardial infarction ECG data - Reduced door-to-balloon and ECG-to-balloon times. Decrease in-hospital and 6-month mortality
    Sveric et al. (2024) [19] AI-driven automated workflows HF and left ventricular ejection fraction estimation Echocardiography images Auto-ECHO: 0.93 Reduced measurement variability (<5%). Improved reliability over traditional methods
    MBS-ECHO: 0.92
    Study AI/ML tool used Disease/symptom Data used Accuracy measure Outcome
    Ippolito et al. (2023) [20] Multi-tasking mask R CNN Pneumonia CXR Accuracy: 93.8% Remarkable accuracy of AI tool found in differentiating between COVID-19 pneumonia, typical bacterial pneumonia, and healthy subjects
    Chagas et al. (2024) [21] Predictive models using Boruta algorithm AIDS- associated PCP LDH, O2 saturation AUC: >0.8 Predictive models differentiated AIDS PCP
    CRP, RR (>24 bpm) dry cough, HIV viral load, CD4 cell count and CXRs
    Salvatore et al. (2021) [22] ResNet- 50 architecture and trained and cross validated COVID-19 pneumonia CXRs AUC: 0.98 AI tools showed potential in correctly differentiating COVID-19 from community acquired pneumonia.
    Carlile et al. (2020) [23] CNN COVID-19 pneumonia CXRS with heat maps - More than 80% physicians found the tool aided in optimizing the workflow and clinical decisions.
    Kwon et al. (2020) [24] DenseNet-121 architecture pretrained on ImageNet COVID-19 pneumonia Chest radiographs and clinical variables AUC: Model improved prediction regarding mortality and intubation requirement.
     0.88 for intubation
     0.82 for mortality
    Lin et al. (2024) [25] Blood culture prediction index and CURB-65 integrated novel Pneumonia CBC, differential leukocyte counts, BUN, age, sex, RR, blood pressure, GCS AUC: 0.713 for cox regression model Integrated models had better predictive performance for low risk patients.
    COX regression model
    Somani et al. (2021) [26] Fusion model- ML model Pulmonary embolism Clinical data and ECGs AUC: 0.84 AUC for fusion model Fusion models performed better than the other models in improving screening of acute PE.
    Choi et al. (2024) [27] Smartphone application ECG BUDDY Pulmonary embolism ECGs AUC: 0.895 Right ventricular dysfunction identified rapidly on ECG as compared to markers like Troponin I and ProBNP.
    QCG-RVDys
    Müller-Peltzer et al. (2021) [28] Computer aided algorithm integrated into Siemens Healthineers software, VB20A Pulmonary embolism CT pulmonary angiography Sensitivity: 47% at lobar and 50% at subsegmental level Computer-aided algorithms lead to higher false positives can be used to assist radiologists but needs radiological examination for confirmation.
    Langius-Wiffen et al. (2023) [29] AI algorithm Pulmonary embolism CT pulmonary angiography Sensitivity: 96.8% Al algorithm showed higher accuracy in detecting pulmonary embolism versus radiologist.
    Specificity: 99.9%
    Savage et al. (2024) [30] AI triage system, the BriefCase for iPE from the vendor Aidoc Pulmonary embolism Contrast enhanced CT scans Sensitivity: 96.2 AI assistance showed increased sensitivity for detecting pulmonary embolism. However, no significant change observed in decreasing time.
    Specificity: 99.9
    Huang et al. (2020) [31] Deep learning 77- layer 3D convolutional neural network (PENet) Pulmonary embolism CT scan AUC: Successful automated application for diagnosis of pulmonary embolism with the AI tool.
    0.84 for internal validation
    0.85 for external validation
    Villacorta et al. (2022) [32] Generalized learning logistic regression using elastic net Pulmonary embolism O2 saturation, history of deep venous thrombosis or pulmonary embolism, immobilization or surgery and D dimer AUC: 0.89 D dimer maintained a low false negative rate
    Hillis et al. (2022) [33] AI model Pneumothorax CXR AUC: Al model detected pneumothorax and tension pneumothorax accurately.
     0.97 for pneumothorax
     0.98 for tension pneumothorax
    Han et al. (2025) [34] Convolutional Recurrent Neural Network and sequence modelling (recurrent neural network) ARDS ECG, HR, RR, SpO2, and non-invasive systolic and diastolic blood pressures, body temperature, AVPU score, age, sex, arterial blood gas analysis AUC: Positive prediction scores and clinical outcomes were established with the AI model.
     0.84 for internal validation
     0.73 for external validation
    Rashid et al. (2022) [35] Multiple AI models ARDS Variable AUC: 0.8–1 AI models reduced cost and improved outcomes amongst ARDS patients.
    Viderman et al. (2024) [36] Feed forward neural networks, CNN and ANN ARDS Variable - AI models shown to perform well in predicting respiratory failure.
    Liong-Rung et al. (2022) [37] GoogleNet Inception V4 architecture Dyspnea CXRs Accuracy: 83.2% High negative predictive value of excluding pulmonary edema & high positive predictive value in diagnosing pneumonia.
    Study AI/ML model Disease Data used Accuracy measure Outcome
    Westwood et al. (2024) [45] Viz LVO, Rapid LVO, Brainomix, Avicenna Stroke Head CTA Sensitivity: 95.4% The study showed that AI software enhanced accuracy in stroke detection.
    Specificity: 79.4%
    Elijovich et al. (2022) [40] Viz AI Stroke Head CTA - The AI software reduced time to notify management team and patients received earlier thrombectomies.
    Gunda et al. (2022) [41] e-Stroke Suite Stroke Non-contrast Head CT and CTA - The AI system helped in decision making process. Reduced time and improved rate of receiving reperfusion treatment strategies.
    Shlobin et al. (2022) [39] CNN, RF and SVM Stroke CTA Sensitivity: 63.6% AI tools accurately triaged and diagnosed patients with large vessel occlusion.
    Specificity: 85.8%
    Paz et al. (2021) [38] Rapid LVO Stroke Head CT scan - The study showed that AI tool had low sensitivity and moderate specificity.
    Amukotuwa et al. (2019) [44] Avicenna CINA Stroke Head CT angiography Sensitivity: 95% Avicenna CINA proved to be one of the most accurate tools in stroke detection.
    Specificity: 79%
    Schmitt et al. (2022) [42] Brainomix e-CTA Stroke Non-contrast enhanced head CT Sensitivity: 0.91 The AI-based algorithm dependably determined if acute intra-cranial hemorrhage were present amongst participants and calculated intra-parenchymal hemorrhage volumes accurately
    Specificity: 0.89
    Wouters et al. (2018) [46] Automated multi-variate imaging model Stroke Diffusion weighted and perfusion weighted imaging AUC: 0.80 The model accurately identified patients of large vessel occlusion within 6-hour window.
    Study AI/ML/DL tool used Disease/symptom Data used Accuracy measure Outcome
    McInnis et al. (2023) [47] AI-EEG Epilepsy EEG AUC: 0.86 Higher diagnostic accuracy than clinical experts
    Tveit et al. (2023) [48] SCORE-AI Epilepsy EEG Accuracy: 88.3% Reduces diagnostic errors and improves interrater agreement
    Keikhosrokiani et al. (2024) [49] AI augmented pathway for digital care pathway for epilepsy (DCPE) Epilepsy Scalp EEG Accuracy: 97.65% for CNN Increase the speed of seizure detection and diagnosis
    Jahan et al. (2023) [51] Cloud computing Epilepsy EEG - Assessment of epileptiform abnormalities and localization of epileptogenic zones
    Seyam et al. (2022) [52] AI-doc Hemorrhage Head CT Accuracy: 93% High diagnostic and predictive value for ICH
    Matsoukas et al. (2022) [53] Logistic regression, support machine vector, random forest, gradient boosting, deep CNN Hemorrhage NCCT and MRI Accuracy: 93.46% for ICH Detected ICH and microbleeds with high accuracy
    Heit et al. (2021) [54] RAPID-ICH Hemorrhage NCCT scans Sensitivity: 0.95 High accuracy in detecting and quantifying the volume of ICH
    Specificity: 0.95
    Wang et al. (2021) [55] GRAD-CAM Hemorrhage CT scans - Improve the accuracy and transparency in ICH detection
    Wismüller et al. (2020) [56] AI-PROBE Hemorrhage CT scans Accuracy: 96.4% TAT reduction for cases of ICH
    Savage et al. (2024) [57] AI-doc (commercial AI triage system) Hemorrhage NCCT Accuracy: 99.2% Diagnostic accuracy and reported TAT with AI assistance was on par with radiologists
    Agarwal et al. (2023) [58] Convolutional neural network, AI-doc, Avicenna.ai Hemorrhage CT and MRI Sensitivity: 0.90 Reflects that most abnormality detection AI studies were not adequately validated in representative clinical cohorts
    Specificity: 0.90
    Davis et al. (2022) [59] AI-doc Hemorrhage CT scans - Reported significant reductions in RTAT along with improved LOS
    Voter et al. (2021) [60] (AI) decision support systems, Aidoc Hemorrhage CT scans Sensitivity: 92.3% Diagnostic performance in cases with a prior history of neurosurgery ICH is weaker
    Specificity: 97.7%
    Ginat et al. (2021) [61] AI-doc Hemorrhage CT - Shortened the time needed to scan reviews and prioritize severe cases
    Hsu et al. (2023) [62] Random forest and logistic regression ALOC due to hyperglycemia Electronic medical records AUC: 0.79 for all-cause mortality Improve the triage process by forecasting patient outcomes
    Obeid et al. (2019) [63] Word2vec Altered mental status Patient notes Accuracy: 94.5% To automate the identification of altered mental status
    El-Rashidy et al. (2023) [64] MIMIC III ALOC EEG data and vital signs Mean absolute error: 0.269 Enhance the predictive and intuitive explanation of consciousness levels
    Study AI/ML tool Disease Data used Accuracy measure Outcome
    Zhou et al. (2022) [65] 47 ML models (ANN, LR, SVM, XGBoost, and RF) AP Clinical, laboratory and imaging parameters - ML models showed potential in diagnosing AP and predicting mortality and recurrence.
    Pan et al. (2024) [66] MSAnet (CNN) AP and AP&PDAC Non-contrast and enhanced abdominal CT AUC: 0.99 AI tools enhanced diagnostic accuracy.
    Kui et al. (2022) [67] EASY-APP (XGBoost) AP RR, body temperature, abdominal muscle reflex, sex, age, and blood glucose levels Accuracy: 89.1% The AI tool predicted risk of severe AP within few hours of hospital admission with high accuracy.
    Podda et al. (2024) [68] ChatGPT Acute biliary pancreatitis Clinical and laboratory values - The AI tool helped optimize the management of patients in ED and ICU.
    İnce et al. (2023) [69] Gradient boost ML algorithm AP Demographic, clinical and laboratory parameters Accuracy: The ML tool proficiently delineated the severity, need of ICU and survival among patients.
     98.25% for ICU need 92.77% for survival
    Park et al. (2020) [70] CNN AA Abdominal CT Accuracy: more than 90% The ML tool diagnosed AA among patient presenting in ED with severe abdominal pain with high accuracy.
    Issaiy et al. (2023) [71] ANN and LR AA Clinical, laboratory and imaging parameters AUC: 0.985 AI tools diagnosed AA and predicted post-surgical risk of sepsis and ICU requirement efficiently.
    Akbulut et al. (2023) [72] CatBoost model Perforated and non-perforated AA CBC, bilirubin, CRP, age and other laboratory values. Accuracy: 92% The AI model distinguished perforated and non-perforated AA with remarkable accuracy.
    Roshanaei et al. (2024) [73] GNB model, RF model, GBA and SVM AA Patient characteristics, laboratory parameters and cause of pain Accuracy: AI tools showed promising results in accurately diagnosing AA in the emergency setting. Gaussian naïve bayes model displayed most superior results.
     95.03% for GNB
     92.55% for RF
     94.41% for GBA
     91.93% for SVM
    Ghareeb et al. (2024) [74] AI platform AA Patient characteristics AUC: 0.97 The AI platform showed to be an effective diagnostic tool in detecting AA.
    Shung at al. (2019) [76] 30 ML models GIB Clinical parameters AUC: ML models showed superiority over clinical tools in predicting mortality in upper GIB. ANN proved to be the most effective model.
     0.93 for ANN
     0.81 for other ML models
    Saraiva et al. (2021) [77] CNN GIB Capsule endoscopy images Accuracy: 99% Model proved to be effective in identifying small bowel lesions and predicting risk of bleeding.
    Mohan et al. (2021) [78] CNN-based CAD GIB or gastrointestinal hemorrhage WCE Accuracy: 95.4% CNN models showed high accuracy, NPV and PPV in diagnosing GIB or hemorrhage by proficiently interpreting WCE images.
    Study AI/ML tool used Disease Data used Accuracy measure Outcome
    Delahanty et al. (2019) [80] Risk of Sepsis score Sepsis Rhee clinical surveillance criteria AUC: 0.93–0.97 ML model proved to be more effective than conventional clinical screening tools.
    Kaji et al. (2019) [82] Long short-term memory recurrent Neural Network Sepsis Clinical parameters AUC: The model effectively predicted daily sepsis occurrence among patients in admitted in ICU.
     0.87 for sepsis
     0.83 for antibiotic administration
    Barton et al. (2019) [79] Supervised gradient -boosted tree model Sepsis Vital signs AUC: 0.88 for sepsis onset The model accurately predicted sepsis 48 hours before onset.
    Mao et al. (2018) [81] InSight Severe sepsis and septic shock Vital signs Accuracy: ML model predicted severe sepsis and septic shock 4 hours before onset with high accuracy.
     0.92 for sepsis
     0.87 for severe sepsis
    Henry et al. (2022) [84] TRWES ML-based warning system Sepsis Clinical parameters - The alert system model decreased time to initiate antibiotic treatment and improved patient outcomes.
    Yuan et al. (2020) [86] Diagnostic algorithm Sepsis Clinical data Accuracy: 82% The AL algorithm can improve patient outcomes by diagnosing sepsis among ICU patients accurately as compared to SOFA.
    XG Boost
    Yan et al. (2022) [87] ML models Sepsis Clinical data notes from providers, demographic, vital signs, laboratory data and medications - The models proved effective in early identification of sepsis.
    Yue et al. (2022) [88] ANN, XGBoost, LR and SVM AKI in sepsis Various clinical and laboratory parameters Accuracy: 0.82 for XGboost The ML models proved to be proficient in predicting AKI among patients with sepsis.
    Zhang et al. (2023) [89] Multiple ML models Mortality in sepsis Various clinical and laboratory parameters Sensitivity: 0.71 The ML models showed superiority in predicting mortality as compared to conventional clinical scoring tools. ML highlighted lactate to be an effective indictor, usually missed by clinical tools.
    Specificity: 0.68
    She et al. (2023) [90] SVM and RF Sepsis Non-targeted liquid chromatography-high-resolution mass spectrometry metabolomics - ML model delineated the association between various metabolites and sepsis that can potentially aid in diagnosis and management.
    Table 1. Overview of studies using AI to diagnose cardiovascular diseases

    AI: artificial intelligence; ML: machine learning; AUC: area under the curve; ECG: electrocardiogram; BMI: body mass index; ARIC: atherosclerosis risk in communities; HF: heart failure; FH: familial hypercholesterolemia; CPH: cox proportional hazards; DLA: deep-learning algorithm; VAE: variational autoencoder; MI: myocardial infarction; ACS: acute coronary syndrome; ECHO: echocardiogram; MBS: medicare benefits schedule echocardiography.

    Table 2. Overview of studies using AI to diagnose respiratory diseases

    AI: artificial intelligence; ML: machine learning; CNN: convolutional neural network; CXR: chest X-ray; COVID-19: coronavirus disease 2019; AIDS: acquired immunodeficiency syndrome; PCP: pneumocystis jirovecii pneumonia; LDH: lactate dehydrogenase; CRP: C-reactive protein; RR: respiratory rate; HIV: human immunodeficiency virus; CD: cluster of differentiation; AUC: area under the curve; CBC: complete blood count; BUN: blood urea nitrogen; GCS: Glasgow coma scale; ECG: electrocardiogram; PE: pulmonary embolism; ProBNP: pro-B-type natriuretic peptide; CT: computed tomography; iPE: incidental pulmonary embolism; 3D: three-dimensional; ARDS: acute respiratory distress syndrome; HR: heart rate; AVPU: alert, verbal, pain, unresponsive; ANN: artificial neural networks.

    Table 3. Overview of studies using AI to diagnose stroke

    AI: artificial intelligence; ML: machine learning; LVO: large vessel occlusion; CTA: computed tomography angiography; CT: computed tomography; CNN: convolutional neural network; RF: random forest; SVM: support vector machine; AUC: area under the curve.

    Table 4. Overview of studies using AI to diagnose neurological diseases

    AI: artificial intelligence; ML: machine learning; DL: deep learning; EEG: electroencephalogram; AUC: area under the curve; CNN: convolutional neural network; CT: computed tomography; ICH: intracranial hemorrhage; NCCT: non-contrast computed tomography; MRI: magnetic resonance imaging; TAT: turnaround time; LOS: length of stay; RTAT: report turnaround time; ALOC: altered level of consciousness.

    Table 5. Overview of studies using AI to diagnose gastro-intestinal diseases

    AI: artificial intelligence; ML: machine learning; ANN: artificial neural network; LR: logistic regression; SVM: support vector machine; RF: random forest; AP: acute pancreatitis; CNN: convoluted neural network; AP&PDAC: acute pancreatitis with pancreatic ductal carcinoma; CT: computed tomography; AUC: area under the curve; RR: respiratory rate; ED: emergency department; ICU: intensive care unit; AA: acute appendicitis; CBC: complete blood count; CRP: C-reactive protein; GNB: Gaussian naïve bayes; GBA: gradient boost algorithm; GIB: gastrointestinal bleeding; ANN: artificial neural network; CAD: computer aided design; WCE: wireless capsule endoscopy NPV: negative predictive value.

    Table 6. Overview of studies using AI to diagnose sepsis

    AI: artificial intelligence; ML: machine learning; AUC: area under the curve; ICU: intensive care unit; SOFA: Sequential Organ Failure Assessment; ANN: artificial neural network; LR: logistic regression; SVM: support vector machine; AKI: acute kidney injury; RF: random forest.

    Table 1.

    Table 2.

    Table 3.

    Table 4.

    Table 5.

    Table 6.

    ACC : Acute and Critical Care
    © The Korean Society of Critical Care Medicine.
    TOP