This repository contains the official implementation of our multimodal deep learning framework for real-time beat-to-beat ΔSBP (delta Systolic Blood Pressure) estimation using physiological signals and demographic data.
We propose a hybrid CNN–BiLSTM–attention model that fuses:
- Photoplethysmogram (PPG)
- Electrocardiogram (ECG)
- Demographic features (e.g., age, gender, BMI)
The model is pretrained on the Aurora BP dataset and fine-tuned on a clinical SCI (Spinal Cord Injury) cohort using SCAI-BP dataset via transfer learning. It satisfies international standards for BP monitoring systems (AAMI, BHS).
If you use this codebase, please cite our paper (coming soon).
├── src/
├── Supplementary_material/
├── README.md
├── LICENSE
└── requirements.txt # Python dependencies