Anticoagulant drugs are commonly prescribed to prevent hypercoagulable states in patients with venous thromboembolism. The choice of the most efficient anticoagulant and the appropriate dosage regimen remain a complex problem because of the intersubject variability in the coagulation kinetics and the effect of blood flow. The rapid assessment of the patient-specific response to anticoagulant regimens would assist clinical decision-making and ensure efficient management of coagulopathy. In this work, we introduce a novel approach that combines computational modeling and deep learning for the fast prediction of the patient-specific response to anticoagulant regimens. We extend a previously developed model to explore the spatio-temporal dynamics of thrombin generation and thrombus formation under anticoagulation therapy. Using a 1D version of the model, we generate a dataset of thrombus formation for thousands of virtual patients by varying key parameters in their physiological range. We use this dataset to train an artificial neural network (ANN) and we use it to predict patient's response to anticoagulant therapy under flow. The algorithm is available and can be accessed through the link: https://github.com/MPS7/ML_coag. It yields an accuracy of 96 % which suggests that its usefulness can be assessed in a randomized clinical trial. The exploration of the model dynamics explains the decisions taken by the algorithm.
Keywords: Anticoagulant; Blood coagulation; Computational fluid dynamics; Personalized medicine; Support vector machines.
Published by Elsevier Inc.