Heart failure classification using deep learning to extract spatiotemporal features from ECG.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Heart failure is a growing global health concern due to population aging.
• Electrocardiogram (ECG) signals contain valuable temporal and spatial characteristics for heart failure assessment.
• The MIMIC-III database provided a rich source for analyzing patient ECG data.

Purpose of the Study:

• To develop a deep learning model for classifying heart failure functional status.
• To leverage ECG signal characteristics for improved heart failure diagnosis.
• To create an automated tool for assisting clinical medical diagnoses.

Main Methods:

• A deep learning model, CNN-LSTM-SE with an attention mechanism, was developed.
• ECG signals were segmented into 2 to 20-second intervals.
• Ablation studies identified optimal 12-second ECG segments for classification.

Main Results:

• The developed model achieved high performance metrics: 99.09% accuracy, 98.99% positive predictive value, 99.03% sensitivity, and 99.65% specificity.
• The model demonstrated superior classification of heart failure using 12-second ECG segments.
• The NYHA functional classification model showed excellent diagnostic capabilities.

Conclusions:

• The proposed deep learning model demonstrates comprehensive performance exceeding similar methods.
• This model can serve as a valuable tool to assist clinicians in diagnosing heart failure.
• The integration of attention mechanisms in ECG analysis shows promise for cardiovascular disease management.