Automated and Interpretable Patient ECG Profiles for Disease Detection, Tracking, and Discovery
View Full Paper
Ask AI
Bookmark
Share
Key result
An automated machine learning pipeline using patient-level ECG profiles accurately detected pulmonary arterial hypertension (AUROC 0.94), hypertrophic cardiomyopathy (AUROC 0.91), and cardiac amyloidosis (AUROC 0.86).
Why the study?
Advances in computing, machine learning, and large-scale data may expand clinical inferences from the ECG while preserving interpretability for medical decision-making.
Do machine learning models using patient-level ECG profiles accurately estimate cardiac structure and detect cardiovascular diseases compared to standard clinical measurements?
Population
36 186 ECGs from the University of California, San Francisco database
Comparison
ECG-profile machine learning models vs clinical measurements
Authors
Discussion
Loading...
Member takes
Implication
May aid ECG-based detection of PAH, HCM, and amyloid; leaves open prospective validation before clinical adoption.
Study Design
Type
Observational (n=36,186)
Multicenter
No
Structured PICO
Do machine learning models using patient-level ECG profiles accurately estimate cardiac structure and detect cardiovascular diseases compared to standard clinical measurements?
P
Population
36,186 ECGs from the University of California, San Francisco database
I
Intervention
Machine learning models (convolutional neural networks and hidden Markov models) using a 725-element patient-level ECG profile
C
Comparator
Standard clinical workflow measurements (141,864 measurements)
O
Outcome
Agreement with clinical estimates for ECG intervals, and discrimination (AUC) for cardiac structure (left ventricular hypertrophy, diastolic dysfunction) and disease detection (pulmonary arterial hypertension, hypertrophic cardiomyopathy, cardiac amyloid, and mitral valve prolapse)surrogate
Main Result
Effect estimate: AUROC 0.94 (95% CI 0.93-0.95)
Machine learning models applied to standard 12-lead ECGs can accurately estimate cardiac structure and detect specific cardiovascular diseases such as pulmonary arterial hypertension and hypertrophic cardiomyopathy.
Limitations
- Models are currently optimized to analyze ECGs in normal sinus rhythm only
- Data derived from a single medical center
Cite This Study
Tison et al. (2019) conducted an observational in Normal sinus rhythm (n=36,186). ecgAI (CNN-HMM derived ECG segmentation and patient-level ECG profiles) vs. Clinical estimates (MUSE/echocardiogram) was evaluated on Detection of pulmonary arterial hypertension (PAH) (AUROC 0.94, 95% CI 0.93-0.95). An automated machine learning pipeline using patient-level ECG profiles accurately detected pulmonary arterial hypertension (AUROC 0.94), hypertrophic cardiomyopathy (AUROC 0.91), and cardiac amyloidosis (AUROC 0.86).
Topics
View Full Paper
Ask AI
Bookmark
Share