By: Shayan Riyaz
The purpose of this project is to explore Heart Rhythms using:
- Signal Processing Techniques
- Machine/Deep Learning
- Causal Analysis/Inference
Currently I am using ~450 subject MIMIC III subject via physionet. I converted Peter Charlton's collate_mimic_perform_af_dataset script for extracting the data from Matlab to python. (Using a mix of LLMs and my own intuiton). Next I used the MSPTDfast peak detection algorithm (again by Peter Charlton & again converted using LLM and my own understanding of the algorithm.).
Datasets Prepared:
-
Training and Validation
- (MIMIC III - Waveform)[https://physionet.org/files/mimic3wdb/1.0/matched/]
- (MIMIC IV - Waveform)[https://physionet.org/files/mimic4wdb/0.1.0/]
- (MIMIC IV ECG - Matched)[https://physionet.org/files/mimic-iv-ecg/1.0/] Mainly notes to classify subject diseases
-
Testing
- CapnoBase
- Currently i created a dummy Conv1D net for peak detection for PPG Signals. (not sure if its necessary but a good learnign exercise).
Goals:
- Noisy Frame Detection using Deep Learning + Simple Statistical Measurments
- Peak Detection using Semi-supervised (pre-labeled using MSPTDfast algorithm) + Reinforcement learning based on physiological behaviour of pleth.
- Calculation of Features such as:
- Time-domain: HRV (Heart Rate Variability), mean HR, PTT (Pulse Transit Time)
- Morphological: Pulse amplitude, rising time, dicrotic notch position
- ? (Still Studying)
- Some Type of Causal Modeling Architecture (X features, treatment) → outcome using TarNet
- Purpose of Causal Modeling: Predicting patient deterioration, alarm risk, personalization
- Charlton, P. H., Prada, E. J. A., Mant, J., & Kyriacou, P. A. (2025). The MSPTDfast photoplethysmography beat detection algorithm: design, benchmarking, and open-source distribution. Physiological Measurement. https://doi.org/10.1088/1361-6579/adb89e
- Charlton, P. H. (n.d.). PPG-Beats. https://ppg-beats.readthedocs.io/en/latest/ AF Annotations
- https://figshare.com/articles/dataset/Atrial_Fibrillation_annotations_of_electrocardiogram_from_MIMIC_III_matched_subset/12149091/1
Dataset Reference:
- Moody, B., Hao, S., Gow, B., Pollard, T., Zong, W., & Mark, R. (2022). MIMIC-IV Waveform Database (version 0.1.0). PhysioNet. https://doi.org/10.13026/a2mw-f949.