A Deep Learning Model for the Identification of Active Contraction Properties of the Myocardium Using Limited Clinical Metrics

Abstract

With the increasing prevalence of cardiovascular diseases globally and limitations in current constitutive models for myocardium mechanics, this study aims to develop a deep learning (DL) model geared toward patient-specific evaluation of left ventricular (LV) myocardium behavior. The primary aim is to bridge the gap between complex constitutive models and their practical clinical applications. We employed DL techniques to construct a model that operates under synthetic clinical metrics and pressure–volume loop data to predict the behavior of the LV myocardium for a whole cardiac cycle. This model generates a waveform of the active contraction parameter and estimates fiber angles at the endocardium and epicardium within a single forward pass. The outputs of our model can be used to infer the unknown parameters in finite element analysis of the myocardium. The DL model demonstrates consistency across various synthetic clinical data scenarios. While the model was tested using both idealized and a single patient-derived geometry, it is important to note that the results have not yet been validated against real clinical data. Overall, the developed DL model offers an innovative approach to understanding and predicting myocardium mechanics in patient-specific clinical settings. By reducing the barrier between the clinical applications and theoretical constitutive models, the proposed method has the potential to support more targeted cardiovascular interventions and diagnostics.