We present a new DNN architecture, SQUWA, for AF detection using PPG data, which includes an innovative attention mechanism. Unlike traditional methods that discard low-quality signals, SQUWA dynamically weighs PPG segments based on their signal quality, directly incorporating this into the AF detection process. This mechanism prioritizes higher-quality segments during prediction and reduces the influence of noisier ones, optimizing the use of data in the overall analysis. Additionally, it processes data points individually rather than as a uniform sample, enhancing detection accuracy and effectiveness. The design principles of SQUWA could also be applied to other fields like human activity and speech recognition, addressing similar issues with noisy data.

First, clone the GitHub repository:

git clone https://github.com/Runz96/SQUWA

To install the core environment dependencies of Raincoat, use environment.yml in config folder:

conda env create -f environment.yml

Confiure the path of training set and validation set in train_adapt.yaml in config folder, training set will not be shared for ethical reasons, except for one publicly accessible

Too train a model:

python train.py

If you find SQUWA useful for your research, please consider citing this paper:

@article{yan2024squwa,
  title={SQUWA: Signal Quality Aware DNN Architecture for Enhanced Accuracy in Atrial Fibrillation Detection from Noisy PPG Signals},
  author={Yan, Runze and Ding, Cheng and Xiao, Ran and Fedorov, Aleksandr and Lee, Randall J and Nahab, Fadi and Hu, Xiao},
  journal={arXiv preprint arXiv:2404.15353},
  year={2024}
}

SQUWA codebase is under MIT license. For individual dataset usage, please refer to the dataset license found in the website.