In the 2024 PhysioNet/CinC Challenge, team "PulsePlex" tackles the dual tasks of reconstructing ECG waveforms from scanned images and classifying ECGs as normal or abnormal. Leveraging image statistics, our waveform reconstruction method synthesizes precise time-series data, yielding an impressive Signal-to-Noise Ratio (SNR) of -18.116, as evidenced by our unofficial submission leaderboard performance. Our method involves extracting mean and standard deviation from images and utilizing them to generate precise waveforms.
For ECG classification, we present a robust deep-learning model based on ResNet50 architecture. We used a Pre-Trained ResNet50 model for Custom training. Through meticulous parameter tuning, including 65 epochs, a learning rate of 0.01, momentum of 0.9, and a weight decay of 0.003, our model achieves exceptional performance. Rigorous experimentation and fine-tuning maximize classification accuracy, resulting in high precision (0.941), recall (0.851), and F1 score (0.894) during cross-validation.
Our integrated approach signifies significant advancements in ECG interpretation, promising implications for clinical diagnosis and disease management. By synergizing innovative algorithms with deep learning techniques, we aim to elevate healthcare practices and enhance patient outcomes.