PMcardio: International Clinical Validations & Research

PMcardio, along with its AI algorithms, has undergone extensive validation through international clinical research encompassing diverse patient groups, and published across various journals. This process has proven its effectiveness in diagnosing 39 cardiovascular diseases including acute coronary occlusion myocardial infarction (OMI), showcasing its precision in interpreting 12-lead ECGs.

1. Validation of an automated artificial intelligence system for 12‑lead ECG interpretation  

Clinical Validations & Medical Journal Publications

• Journal: Journal of Electrocardiology
• Topics: Core Diagnostics, Artificial Intelligence
• Date: 12/23/2023
• Type: Study

This validation study evaluated the performance of an artificial intelligence (AI)-based electrocardiogram (ECG) interpretation system as a diagnostic tool. The study describes developing and validating an AI system consisting of six deep neural networks trained on over 900,000 ECGs to discern 20 key diagnostic patterns. Diagnostic accuracy was benchmarked against an independent test set composed of ECGs annotated by expert cardiologists. Key performance metrics included sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and F1 score.

With over 932,711 ECGs drawn from 173,949 patients to develop the AI system, 11,932 annotated ECG labels from the independent test set were used for comparison. The AI system demonstrated high sensitivity, specificity, and positive and negative predictive values across all diagnostic categories. The AI system outperformed the current computerized interpretation of ECGs (CIE) across 13 of the 20 key diagnostic patterns while matching the CIE’s performance for the other 7 diagnoses. Notably, the model’s ability to identify atrial fibrillation was near perfect. These findings indicate the strong potential for AI-powered ECG interpretation to enhance diagnostic accuracy in clinical environments and serve as a valuable tool for healthcare workers.

Read more in the Journal of Electrocardiology

2. International evaluation of an artificial intelligence-powered ECG model detecting acute coronary occlusion myocardial infarction

• Journal: European Heart Journal – Digital Health
• Topic: Artificial Intelligence, OMI AI ECG Model, Acute Coronary Syndrome
• Type: Study
• Date: 11/28/2023

This study aimed to develop and externally validate an AI model that can detect Occlusion Myocardial Infarction (OMI) using standard 12-lead ECGs, given that one-third of Non-ST-elevation myocardial infarction (NSTEMI) patients have OMIs that are often associated with poor outcomes due to delayed treatment.

European Heart Journal: Clinical Validation Study of the PMcardio OMI AI Model

The AI model, which was trained on a diverse data set of 18,616 ECGs and tested on an independent holdout set, demonstrated robust performance with an AUROC of 0.941, a significant improvement over existing ST-elevation Myocardial Infarction (STEMI) criteria. These findings indicate the notable potential of AI in improving the detection of acute OMI, suggesting an opportunity to enhance patient triage and timely intervention for ACS patients.

Read more in the European Hearth Journal – Digital Health

3. Diagnostic accuracy of a smartphone application for artificial intelligence-based interpretation of 12-lead ECG in primary care (AMSTELHEART-1) 

• Journal: European Heart Journal
• Topic: Artificial Intelligence, Core Diagnostics, External
• Type: Abstract
• Date: 11/9/2023

This external retrospective validation study found that PMcardio is an accurate and reliable tool for diagnosing major ECG abnormalities and has nearly perfect properties for detecting atrial fibrillation in the primary care setting. The performance of the app was comparable on Android and iOS platforms.

Read more in the European Heart Journal

4. Validation of an artificial intelligence model for 12-lead ECG interpretation

• Journal: European Heart Journal
• Topic: Artificial Intelligence, Core Diagnostics
• Type: Abstract
• Date: 11/9/2023

This study assessed the accuracy of an AI-powered model in interpreting standard 12-lead ECGs, benchmarking its performance against primary care physicians and cardiologists. Using a Deep Neural Network (DNN) trained on over 900,000 ECGs, the model was designed to detect 38 common electrocardiographic abnormalities grouped into six categories.

The model’s predictive performance, evaluated on an independent test set, was found to surpass the accuracy of primary care physicians and was non-inferior to cardiologists. The AI model demonstrated high sensitivity, specificity, and positive and negative predictive values across all diagnostic categories. Notably, the model’s ability to identify atrial fibrillation was near perfect. The findings from this study highlight the potential of this AI-powered ECG model as a precise and accessible diagnostic tool for healthcare professionals.

Read more in the European Heart Journal

5. Validation of Deep Learning System for Comprehensive 12-Lead ECG Interpretation 

• Journal: Circulation AHA Journal
• Topic: Artificial Intelligence, Core Diagnostics
• Type: Abstract
• Date: 11/6/2023

This study assessed the accuracy of an AI-powered model in interpreting standard 12-lead ECGs, benchmarking its performance against primary care physicians and cardiologists. Using a Deep Neural Network (DNN) trained on over 900,000 ECGs, the model was designed to detect 38 common electrocardiographic abnormalities grouped into six categories.

The model’s predictive performance, evaluated on an independent test set, was found to surpass the accuracy of primary care physicians and was non-inferior to cardiologists. The AI model demonstrated high sensitivity, specificity, and positive and negative predictive values across all diagnostic categories. Notably, the model’s ability to identify atrial fibrillation was near perfect. The findings from this study highlight the potential of this AI-powered ECG model as a precise and accessible diagnostic tool for healthcare professionals.

Read more in the at Circulation AHA Journal

6. Deep Learning Electrocardiogram Detecting Acute Coronary Occlusion in Myocardial Infarction Presenting Without ST-Elevation

• Journal: Circulation AHA Journal
• Topic: Acute Coronary Syndromes, Artificial Intelligence, OMI AI ECG Model
• Type: Abstract
• Date: November 6, 2023

This study aimed to develop and externally validate an AI model that can detect occlusion myocardial infarction (OMI) using standard 12-lead ECGs, given that one-third of Non-ST-elevation myocardial infarction (NSTEMI) patients have OMIs that are often associated with poor outcomes due to delayed treatment.

The AI model, which was trained on a diverse data set of 18,616 ECGs and tested on an independent holdout set, demonstrated robust performance with an AUROC of 0.941, a significant improvement over existing ST-elevation Myocardial Infarction (STEMI) criteria.

These findings indicate the notable potential of AI in improving the detection of acute OMI, suggesting an opportunity to enhance patient triage and timely intervention for ACS patients.

Read more in the Circulation AHA Journal

7. Poor prognosis of total culprit artery occlusion in patients presenting with NSTEMITopics

• Journal: European Heart Journal
• Topic: Acute Coronary Syndromes, Big Data
• Type: Abstract
• Date: August 29, 2023

This retrospective single-center study compared clinical presentation, management, and all-cause mortality among patients with ST-Elevation Myocardial Infarction (STEMI), Non-ST Elevation Myocardial Infarction with occlusion (NSTEMI-OMI), and Non-ST Elevation Myocardial Infarction without occlusion (NSTEMI-NOMI). A total of 9,943 patients were classified and analyzed.

The results showed that while patients with NSTEMI-OMI and STEMI had similar baseline demographics and angiographic findings, NSTEMI-OMI patients experienced longer delays to reperfusion and had a higher short and long-term all-cause mortality rate compared to STEMI patients. This outcome underscores the need to improve timely identification for NSTEMI-OMI.

Read more in the European Heart Journal

8. ECG-based deep learning for detecting epicardial coronary occlusion in acute myocardial infarction

• Journal: European Heart Journal
• Topic: OMI AI ECG Model, Acute Coronary Syndrome, Artificial Intelligence
• Type: Abstract
• Date: August 29, 2023

This study aimed to develop and externally validate an AI model that can detect occlusion myocardial infarction (OMI) using standard 12-lead ECGs, given that one-third of Non-ST-elevation myocardial infarction (NSTEMI) patients have OMIs that are often associated with poor outcomes due to delayed treatment.

The AI model, which was trained on a diverse data set of 18,616 ECGs and tested on an independent holdout set, demonstrated robust performance with an AUROC of 0.941, a significant improvement over existing ST-elevation Myocardial Infarction (STEMI) criteria. These findings indicate the notable potential of AI in improving the detection of acute OMI, suggesting an opportunity to enhance patient triage and timely intervention for ACS patients.

Read more at European Heart Journal

9. Poor outcomes in occlusive ACS patients presenting without ST-elevation

• EuroPCR
• Topic: Acute Coronary Syndromes, Artificial Intelligence, OMI AI ECG Model
• Type: Abstract
• Date: May 19, 2023

This retrospective single-center study compared clinical presentation, management, and all-cause mortality among patients with ST-Elevation Myocardial Infarction (STEMI), Non-ST Elevation Myocardial Infarction with occlusion (NSTEMI-OMI), and Non-ST Elevation Myocardial Infarction without occlusion (NSTEMI-NOMI). A total of 9,943 patients were classified and analyzed.

The results showed that while patients with NSTEMI-OMI and STEMI had similar baseline demographics and angiographic findings, NSTEMI-OMI patients experienced longer delays to reperfusion and had a higher short and long-term all-cause mortality rate compared to STEMI patients. This outcome underscores the need to improve timely identification for NSTEMI-OMI.

Read more at EuroPCR

10. Diagnostic accuracy of the PMcardio smartphone application for artificial intelligence–based interpretation of electrocardiograms in primary care (AMSTELHEART-1)  

• Journal: Cardiovascular Digital Health Journal
• Topic: Artificial Intelligence, Core Diagnostics, External
• Type: Study
• Date: April 5, 2023

This external retrospective validation study found that PMcardio is an accurate and reliable tool for diagnosing major ECG abnormalities and has nearly perfect properties for detection of atrial fibrillation in the primary care setting. The performance of the app was comparable on Android and iOS platforms.

Read the abstract here

11. Revolutionizing ECG Interpretation with AI-powered universal smartphone technology

• Journal: Journal of Electrocardiology
• Topic: Artificial Intelligence, Core Diagnostics
• Type: Abstract
• Date: March 30, 2023

Computerized interpretation of electrocardiograms (CIE) has been developed to improve diagnoses worldwide. However, obstacles such as noise, access to physician-validated ECG data, and lack of criteria remain. We present a certified medical device utilizing AI to standardize and digitize ECG images, interpret individual cardio diseases, and provide recommendations.

Read more at Journal of Electrocardiology

12. Validation of an automated deep neural network for 12-lead electrocardiogram diagnosis in primary care

• Journal: Journal of Electrocardiology
• Topic: Artificial Intelligence, Core Diagnostics
• Type: Abstract
• Date: March 30, 2023

This study demonstrated that a deep neural network-based algorithm yielded accurate diagnoses of 38 cardiovascular diagnoses with significantly higher performance than primary care physicians. The AI-powered EKG algorithm yielded a mean sensitivity of 0.930 (0.911–0.949), specificity of 0.967 (0.952–0.981), PPV of 0.950 (0.932–0.967), NPV of 0.966 (0.951–0.981), and MCC of 0.906 (0.888–0.921).

Read more at Journal of Electrocardiology

13. High epicardial fat volume is associated with atrial fibrillation recurrences after catheter ablation

• Journal: European Heart Journal
• Topic: Artificial Intelligence, Atrial Fibrillation
• Type: Abstract
• Date: October 3, 2022

This retrospective study of 92 patients with paroxysmal AF undergoing catheter ablation aiming at PVI suggested that CT-quantified EFT volume is significantly associated with AF recurrence at 1-year post ablation, suggesting Epicardial Adipose Tissue to be an adverse pro-fibrillatory factor.

Read more at European Heart Journal

14. Predictors of long-term atrial fibrillation recurrence after catheter ablation: non-linear analytical approach for individualized prognostic stratification

• Journal: European Heart Journal
• Topic: Artificial Intelligence, Atrial Fibrillation
• Type: Abstract
• Date: October 3, 2022

This study demonstrated an association between higher CT-quantified epicardial fat volume and recurrence of atrial fibrillation and the potential of machine learning models to create more individualized outcomes for catheter ablation of pulmonary vein isolation in patients with paroxysmal atrial fibrillation.

Read the abstract here

15. Utilizing longitudinal data in assessing all-cause mortality in patients hospitalized with heart failure

• Journal: ESC Heart Failure Journal
• Topic: Artificial Intelligence, Big Data, Heart Failure
• Type: Study
• Date: June 13, 2022

A machine learning-based predictive algorithm for all-cause mortality in a cohort of patients hospitalized with new onset or worsened heart failure was developed utilizing a large time-series dataset to capture the entirety of patient management from initial contact to long-term follow-up. The system had robust predictive performance across different heart failure phenotypes and was able to consider the evolution of data points over time to provide individualized prognosis at patient contacts.

Read more at ESC Heart Failure

16. Unbiased Deep Learning Approach Utilizing Longitudinal Data in Assessing All-Cause Mortality in Patients With a De Novo or Worsened Heart Failure

• Journal: Circulation AHA Journal
• Topic: Artificial Intelligence, Big Data, Heart Failure
• Type: Abstract
• Date: November 8, 2021

More information: Electronic patient records were used to develop an artificial intelligence (AI) model to predict all-cause mortality in heart failure patients, yielding superior performance compared to standard clinical scores. These results suggest AI models have the potential to be used to guide clinical risk stratification in a point-of-care approach.

Read more at Circulation

17. Deep learning for mortality prediction in patients with a de-novo or worsened heart failure

• Journal: European Heart Journal
• Topic: Artificial Intelligence, Big Data, Heart Failure
• Type: Abstract
• Date: November 8, 2021

This study used deep artificial intelligence and an unbiased predictive algorithm to suggest the potential of AI-based predictive models in risk stratification for all-cause mortality in heart failure, with robust accuracy across its phenotypes.

Read more at European Heart Journal