Graph Neural Network based Future Clinical Events Prediction from Invasive Coronary Angiography

Early prediction of future clinical events from invasive coronary angiography (ICA) remains a daily challenge in clinical routine practice. In this study, we hypothesize that stenosis's geometry information could benefit the prediction of future events from ICA. To address this, we propose a framework that employs graph neural networks (GNNs) to exploit geometry information from ICA and integrates it with clinical information to predict the occurrence of events at the stenosis level. The proposed model can be extended to predict events using two-view imaging data as well. The performance is compared to classical baseline models on a dataset comprising 1551 stenosis, out of which 414 exhibited an event in the following two years. The results illustrate that the proposed approach outperforms other models, with F1-scores of 0.57 and 0.59 for one-view and two-view data, respectively. To the best of our knowledge, this is the first work that investigates the importance of the geometry information for future events prediction in a learning context. The code is available at https://github.com/xsunn/eventsPre.