Projects
All Projects
Selected projects in cardiac monitoring, efficient deep learning, and open-source biosignal processing.
6 / 6 projects
BioSPPy
Built BioSPPy’s HRV module and EDF reader; maintain its APIs, documentation and plotting system.
- Result
- 1M+ downloads · 470+ citations · SoftwareX 2024
PythonSignal Processing
Frugal AI for Automated Defibrillation
Developed compact ECG models and their evaluation; worked with Caroline Stehlé on INT8 quantization and STM32 implementation.
- Result
- INT8 weights: 408 → 102 KiB. FP32: 99.7% sensitivity / 99.3% specificity on held-out 6-second ECGs.
Longitudinal AFib Prediction
Designed and implemented SurviFormer end to end: longitudinal ECG survival modelling, training and external validation.
- Result
- 140k patients · 517k ECGs in CODE. External C-index: 0.805.
CardioLeather
Built an ECG processing pipeline and LSTM models for driver drowsiness detection in collaboration with CardioID.
- Result
- 20-RR input windows · DROZY dataset · 5-fold cross-validation
YOUR-Lead
Paper ECG digitization using synthetic training images, U-Net trace segmentation, and YOLOv8 lead detection. Developed for the PhysioNet Challenge 2024.
Automated ARI Extraction
Wavelet-based extraction of activation-recovery intervals from intracardiac electrograms, guided by surface ECG timing and evaluated across rhythms and tissue types in an infarcted swine cohort.