Rapid thrombogenesis prediction in COVID-19 patients using DNN with data labeling

Abstract

The rapid spread of COVID-19 globally has led to a surge in patients, among whom the incidence of blood clotting has been observed. This phenomenon poses a significant risk, potentially leading to cerebral hemorrhage and severe complications. Early detection of such blood clotting events is crucial for timely intervention and management of secondary complications. Computational Fluid Dynamics (CFD) simulations offer a means to detect these clotting phenomena, albeit with a drawback of time-intensive confirmation, spanning from several days to months. To address this challenge, we propose a framework for early detection of blood clotting events utilizing a deep neural network model. Leveraging patient-specific data derived from CFD simulations, our model learns to predict clotting events, expediting the diagnostic process. Hyperparameter tuning was employed to optimize the deep neural network algorithm for enhanced accuracy and performance. Furthermore, our framework facilitated comparative simulations across various machine learning algorithms, aiding in the identification of optimal data sizes for achieving high predictive accuracy. By combining computational modeling with machine learning techniques, our approach offers a promising avenue for expedited detection and management of thrombogenesis phenomena in COVID-19 patients.