Wearable Real Time Blood Pressure Prediction Using Mobile ECG, PPG Data

• ResNet18-based CNN architecture for feature extraction from time-series biosignals (ECG & PPG)

• Attention mechanism to fuse ECG and PPG features for enhanced prediction

• Real-time prediction capability using 5-second overlapping signal windows

• Continuous systolic and diastolic signal prediction, not just single-value estimates

• Data augmentation techniques (random baseline and time offsets) to improve generalization

• Custom dataset collected from healthy and pathological patients using noninvasive sensors

•  MSE loss function for training with optimized hyperparameters using SGD