PiLA: PPG-based Cardiac Disease Diagnosis Model

Overview

This repository provides pretrained and finetuned model weights for PiLA, a deep learning framework for cardiac disease diagnosis from photoplethysmography (PPG) signals.

The model supports multimodal physiological inputs, including:

• PPG
• VPG (velocity PPG)
• APG (acceleration PPG)

Files

• pretrain_model.pth: self-supervised pretrained weights
• model_best.pth: finetuned model for downstream diagnosis

Code

Source code is available at: https://github.com/EZAIJ/PiLA-PPG

Data

Due to privacy and ethical constraints, the dataset is not publicly available.

Notes

• The released weights do not contain any identifiable or participant-level data.
• This model is provided for research use only.

Paper

https://arxiv.org/abs/2602.04266