This study developed an interpretable deep learning model to enhance the accuracy and interpretability of ECG interpretation, aiming to advance cardiovascular diagnosis and mortality risk stratification. The model was trained on a large dataset of 2.3 million ECG recordings from 1.6 million subjects and validated across four medical tasks: arrhythmia diagnosis, gender identification, hypertension screening, and mortality risk stratification. The model demonstrated cardiologist-level accuracy in interpreting ECGs, identifying lead-specific and disease-specific morphological features, and providing personalized healthcare. Key findings include: 1. **Arrhythmia Diagnosis**: The model achieved high accuracy (AUC score of 0.998) and F1 score of 0.948 in identifying various arrhythmias, outperforming junior professionals and a state-of-the-art benchmark. 2. **Gender Identification**: The model accurately identified gender with an AUC score of 0.964, showing higher performance in younger age groups. 3. **Hypertension Screening**: The model effectively screened for hypertension with an AUC score of 0.839, achieving comparable results to previous studies using multiple leads. 4. **Mortality Risk Stratification**: The ECG-predicted results were used to stratify mortality risk, demonstrating comparable performance to raw ECG data in predicting mortality rates. The study also identified dominant ECG leads for each task, providing insights into the importance of specific leads in different conditions. The model's interpretability and performance suggest its potential for developing more efficient and accurate wearable devices for cardiac monitoring and healthcare applications.This study developed an interpretable deep learning model to enhance the accuracy and interpretability of ECG interpretation, aiming to advance cardiovascular diagnosis and mortality risk stratification. The model was trained on a large dataset of 2.3 million ECG recordings from 1.6 million subjects and validated across four medical tasks: arrhythmia diagnosis, gender identification, hypertension screening, and mortality risk stratification. The model demonstrated cardiologist-level accuracy in interpreting ECGs, identifying lead-specific and disease-specific morphological features, and providing personalized healthcare. Key findings include: 1. **Arrhythmia Diagnosis**: The model achieved high accuracy (AUC score of 0.998) and F1 score of 0.948 in identifying various arrhythmias, outperforming junior professionals and a state-of-the-art benchmark. 2. **Gender Identification**: The model accurately identified gender with an AUC score of 0.964, showing higher performance in younger age groups. 3. **Hypertension Screening**: The model effectively screened for hypertension with an AUC score of 0.839, achieving comparable results to previous studies using multiple leads. 4. **Mortality Risk Stratification**: The ECG-predicted results were used to stratify mortality risk, demonstrating comparable performance to raw ECG data in predicting mortality rates. The study also identified dominant ECG leads for each task, providing insights into the importance of specific leads in different conditions. The model's interpretability and performance suggest its potential for developing more efficient and accurate wearable devices for cardiac monitoring and healthcare applications.