Deep Spatio-Temporal Multiplex Graph Learning for Cardiac Imaging Classification

Jaume Banus; Augustin Ogier; Roger Hullin; Philippe Meyer; Ruud B. van Heeswijk; Jonas Richiardi

ICASSP 2023accepted0 citations

Deep Spatio-Temporal Multiplex Graph Learning for Cardiac Imaging Classification

Jaume Banus, Augustin Ogier, Roger Hullin, Philippe Meyer, Ruud B. van Heeswijk, Jonas Richiardi

Abstract

Cardiovascular diseases are the leading cause of mortality worldwide. Cardiac imaging is critical for precise characterization of cardiac structure and function, and is key in diagnosis, therapeutic management, and prognosis. In this work, we propose a novel representation of spatio-temporal cardiac data as a multiplex graph and develop a multi-level message passing neural network to classify clinical groups corresponding to different cardiovascular diseases. Using open data from the Automated Cardiac Diagnosis Challenge (N=150), our results show that the multiplex representation extracts discriminative features from the data (up to 94% accuracy compared to up to 80% for well-tuned baselines such as XGboost and MLP). Ablation studies show the bias induced by the graph’s spatio-temporal structure improves generalization.

BibTeX

@inproceedings{icassp2023_deepspatiotempor,
  title = {Deep Spatio-Temporal Multiplex Graph Learning for Cardiac Imaging Classification},
  author = {Jaume Banus and Augustin Ogier and Roger Hullin and Philippe Meyer and Ruud B. van Heeswijk and Jonas Richiardi},
  booktitle = {ICASSP 2023},
  year = {2023}
}