Journal: Journal of the American College of Emergency Physicians Open

Article Title: Predicting Occlusion Myocardial Infarctions in the Emergency Department Using Artificial Intelligence

doi: 10.1016/j.acepjo.2025.100299

Figure Lengend Snippet: Exclusion criteria and data split. Exclusion criteria and data split. Each patient is only represented once. The random split was stratified on the outcome (OMI), so that each set contained the same proportion of OMI cases. ECG, electrocardiogram; ED, emergency department; hs-cTnT, high-sensitivity cardiac troponin T; OMI, occlusion myocardial infarction.

Article Snippet: By comparison, Al-Zaiti et al evaluated a commercial ECG interpretation system (Philips DXL diagnostic algorithm) and found a sensitivity of 79% and specificity of 80% for OMI in patients with chest pain with ECGs from ambulances.

Techniques:

Journal: Journal of the American College of Emergency Physicians Open

Article Title: Predicting Occlusion Myocardial Infarctions in the Emergency Department Using Artificial Intelligence

doi: 10.1016/j.acepjo.2025.100299

Figure Lengend Snippet: Summary of the occlusion myocardial infarction (OMI) annotation process. Overview of the OMI annotation process. Uni-G STEMI refers to “STEMI equivalent” diagnoses from the University of Glasgow ECG analysis program by MacFarlane et al. CABG, coronary artery bypass graft surgery; CAG, coronary angiography; hs-cTnT, high-sensitivity cardiac troponin T; NOMI, negative occlusion myocardial infarction; OMI, occlusion myocardial infarction; PCI, percutaneous coronary intervention; STEMI, ST-segment elevation myocardial infarction; TIMI, thrombolysis in myocardial infarction.

Article Snippet: By comparison, Al-Zaiti et al evaluated a commercial ECG interpretation system (Philips DXL diagnostic algorithm) and found a sensitivity of 79% and specificity of 80% for OMI in patients with chest pain with ECGs from ambulances.

Techniques:

Journal: Journal of the American College of Emergency Physicians Open

Article Title: Predicting Occlusion Myocardial Infarctions in the Emergency Department Using Artificial Intelligence

doi: 10.1016/j.acepjo.2025.100299

Figure Lengend Snippet: Model structure. The boxes on the left are the feature groups, with the numbers on each arrow indicating their size and shape. The arrow on the left indicates the relative time at which the different feature groups are typically available at the emergency department (ED). POC refers to point-of-care blood samples. CAT means concatenation. Only the 8 linearly independent electrocardiogram (ECG) leads were used as input to the ResNet model. AI, artificial intelligence; CAT, concatenation; ED, emergency department; hs-cTnT, high-sensitivity cardiac troponin T; OMI, occlusion myocardial infarction.

Article Snippet: By comparison, Al-Zaiti et al evaluated a commercial ECG interpretation system (Philips DXL diagnostic algorithm) and found a sensitivity of 79% and specificity of 80% for OMI in patients with chest pain with ECGs from ambulances.

Techniques:

Journal: Journal of the American College of Emergency Physicians Open

Article Title: Predicting Occlusion Myocardial Infarctions in the Emergency Department Using Artificial Intelligence

doi: 10.1016/j.acepjo.2025.100299

Figure Lengend Snippet: Receiver operating characteristic (ROC) curves for OMI classification. ROC curves for the AI model evaluated on the validation set using different inputs that correspond to when new information becomes available at the emergency department. The positive class is OMI; the negative class is everyone else. Also plotted are the STEMI criteria and the Uni-G algorithm statements corresponding to STEMI. The dashed line corresponds to randomly guessing the outcome. AUC, area under the ROC curve; ECG, electrocardiogram; hs-cTnT, high-sensitivity cardiac troponin T; OMI, occlusion myocardial infarction; POC, point-of-care blood samples; ROC, receiver operating characteristic; STEMI, ST-segment elevation myocardial infarction.

Article Snippet: By comparison, Al-Zaiti et al evaluated a commercial ECG interpretation system (Philips DXL diagnostic algorithm) and found a sensitivity of 79% and specificity of 80% for OMI in patients with chest pain with ECGs from ambulances.

Techniques: Biomarker Discovery

Journal: JACC: Advances

Article Title: Performance of Single-Lead Handheld Electrocardiograms for Atrial Fibrillation Screening in Primary Care

doi: 10.1016/j.jacadv.2023.100616

Figure Lengend Snippet: Study Overview and Predictive Values of AliveCor Algorithm Interpretations (A) An overview of the current study. A total of 38,190 handheld 1L ECG tracings were generated in the context of screening within the VITAL-AF trial. Of these, we excluded tracings performed among individuals with known AF at the time of screening, as well as those overread as uninterpretable by cardiologist readers, resulting in 31,376 tracings in the primary analysis (see text). We quantified the test characteristics of the AliveCor 1L ECG algorithm against cardiologist overread (primary analysis), against electrophysiologist overread (secondary analysis), and against clinical interpretation of a 12-lead ECG performed on the same day as screening (secondary analysis). (B) The positive predictive values (PPVs) (red) and negative predictive values (NPVs) (green) for the 4 AliveCor automated 1L ECG interpretations (x-axis), using cardiologist overread as the gold standard. Only PPV is depicted for possible AF , which always denotes a positive test, while only NPV is depicted for normal which always denotes a negative test. Since unclassified and no analysis represent equivocal findings, both PPV and NPV are shown. AF = atrial fibrillation; ECG = electrocardiogram.

Article Snippet: We quantified the test characteristics of the AliveCor 1L ECG algorithm against cardiologist overread (primary analysis), against electrophysiologist overread (secondary analysis), and against clinical interpretation of a 12-lead ECG performed on the same day as screening (secondary analysis). (B) The positive predictive values (PPVs) (red) and negative predictive values (NPVs) (green) for the 4 AliveCor automated 1L ECG interpretations (x-axis), using cardiologist overread as the gold standard.

Techniques: Generated

Journal: JACC: Advances

Article Title: Performance of Single-Lead Handheld Electrocardiograms for Atrial Fibrillation Screening in Primary Care

doi: 10.1016/j.jacadv.2023.100616

Figure Lengend Snippet: Test Characteristics of AliveCor Algorithm Interpretations of 1L ECG Tracings Using Cardiologist Overread as Reference Overall and Stratified by Age

Article Snippet: We quantified the test characteristics of the AliveCor 1L ECG algorithm against cardiologist overread (primary analysis), against electrophysiologist overread (secondary analysis), and against clinical interpretation of a 12-lead ECG performed on the same day as screening (secondary analysis). (B) The positive predictive values (PPVs) (red) and negative predictive values (NPVs) (green) for the 4 AliveCor automated 1L ECG interpretations (x-axis), using cardiologist overread as the gold standard.

Techniques:

Journal: JACC: Advances

Article Title: Performance of Single-Lead Handheld Electrocardiograms for Atrial Fibrillation Screening in Primary Care

doi: 10.1016/j.jacadv.2023.100616

Figure Lengend Snippet: Effect of Unclassified and No Analysis Tracings on 1L ECG Diagnostic Performance Stratified by Age of Screening Depicted are the test characteristics (sensitivity, top left; positive predictive value [PPV], top right; specificity, bottom left, negative predictive value [NPV], bottom right) of the AliveCor 1L ECG algorithm using cardiologist overread as the gold standard. In each plot, the relevant metric is plotted across age group (X-axis), and stratified by whether equivocal tracings (ie, unclassified or no analysis ) are considered positive vs negative results. In each plot, the hashed horizontal line depicts the corresponding estimate in the overall sample. ECG = electrocardiogram.

Article Snippet: We quantified the test characteristics of the AliveCor 1L ECG algorithm against cardiologist overread (primary analysis), against electrophysiologist overread (secondary analysis), and against clinical interpretation of a 12-lead ECG performed on the same day as screening (secondary analysis). (B) The positive predictive values (PPVs) (red) and negative predictive values (NPVs) (green) for the 4 AliveCor automated 1L ECG interpretations (x-axis), using cardiologist overread as the gold standard.

Techniques: Diagnostic Assay

Journal: JACC: Advances

Article Title: Performance of Single-Lead Handheld Electrocardiograms for Atrial Fibrillation Screening in Primary Care

doi: 10.1016/j.jacadv.2023.100616

Figure Lengend Snippet: Study Overview and Predictive Values of AliveCor Algorithm Interpretations (A) An overview of the current study. A total of 38,190 handheld 1L ECG tracings were generated in the context of screening within the VITAL-AF trial. Of these, we excluded tracings performed among individuals with known AF at the time of screening, as well as those overread as uninterpretable by cardiologist readers, resulting in 31,376 tracings in the primary analysis (see text). We quantified the test characteristics of the AliveCor 1L ECG algorithm against cardiologist overread (primary analysis), against electrophysiologist overread (secondary analysis), and against clinical interpretation of a 12-lead ECG performed on the same day as screening (secondary analysis). (B) The positive predictive values (PPVs) (red) and negative predictive values (NPVs) (green) for the 4 AliveCor automated 1L ECG interpretations (x-axis), using cardiologist overread as the gold standard. Only PPV is depicted for possible AF , which always denotes a positive test, while only NPV is depicted for normal which always denotes a negative test. Since unclassified and no analysis represent equivocal findings, both PPV and NPV are shown. AF = atrial fibrillation; ECG = electrocardiogram.

Article Snippet: To quantify the performance of the 1L ECG automated algorithm interpretation for detecting AF, we calculated the positive predictive value (PPV) and the negative predictive value (NPV) for each AliveCor reading.

Techniques: Generated

Journal: JACC: Advances

Article Title: Performance of Single-Lead Handheld Electrocardiograms for Atrial Fibrillation Screening in Primary Care

doi: 10.1016/j.jacadv.2023.100616

Figure Lengend Snippet: Effect of Unclassified and No Analysis Tracings on 1L ECG Diagnostic Performance Stratified by Age of Screening Depicted are the test characteristics (sensitivity, top left; positive predictive value [PPV], top right; specificity, bottom left, negative predictive value [NPV], bottom right) of the AliveCor 1L ECG algorithm using cardiologist overread as the gold standard. In each plot, the relevant metric is plotted across age group (X-axis), and stratified by whether equivocal tracings (ie, unclassified or no analysis ) are considered positive vs negative results. In each plot, the hashed horizontal line depicts the corresponding estimate in the overall sample. ECG = electrocardiogram.

Article Snippet: To quantify the performance of the 1L ECG automated algorithm interpretation for detecting AF, we calculated the positive predictive value (PPV) and the negative predictive value (NPV) for each AliveCor reading.

Techniques: Diagnostic Assay

Journal: Journal of Personalized Medicine

Article Title: A Deep Learning Algorithm for Detecting Acute Pericarditis by Electrocardiogram

doi: 10.3390/jpm12071150

Figure Lengend Snippet: Summary of model performance as the area under the receiver operating characteristic curve for predicting pericarditis. The ROC curves were made by the predictions of the deep learning model (DLM) using raw ECG signals and the XGB model integrating ECG measures (8 numerical values and 31 diagnostic labels), respectively. Each point represents the performance of humans and Philips automatic ECG interpretation. The cut points of the DLM and XGB model were based on Youden’s index in the tuning set.

Article Snippet: The AUCs were 0.954 and 0.952 in the validation dataset and chest pain subset, respectively, with the same sensitivities of 78.9% and similar specificities of 97.7% and 97.6%, which were significantly better than the XGB model and Philips automatic ECG interpretation.

Techniques: Diagnostic Assay

Journal: Journal of Personalized Medicine

Article Title: A Deep Learning Algorithm for Detecting Acute Pericarditis by Electrocardiogram

doi: 10.3390/jpm12071150

Figure Lengend Snippet: 3-day CV- and non-CV-caused hospitalization in non-pericarditis cases stratified by DLM classification. DLM identification was defined as the intersection of DLM-pericarditis and DLM-STEMI. A higher risk of 3-day CV-caused hospitalization was present when the DLM defined the ECG as abnormal compared with those who were classified as having a normal ECG by DLM. The numbers reported in the legend are the hazard ratios.

Article Snippet: The AUCs were 0.954 and 0.952 in the validation dataset and chest pain subset, respectively, with the same sensitivities of 78.9% and similar specificities of 97.7% and 97.6%, which were significantly better than the XGB model and Philips automatic ECG interpretation.

Techniques: