A Multimodal Machine Learning Approach to Omics-Based Risk Stratification in Coronary Artery Disease

摘要

This study aims at developing a personalized model for coronary artery disease (CAD) risk stratification based on machine learning modelling of non-imaging data, i.e. clinical, molecular, cellular, inflammatory, and omics data. A multimodal architectural approach is proposed whose generalization capability, with respect to CAD stratification, is currently evaluated. Different data fusion techniques are investigated, ranging from early to late integration methods, aiming at designing a predictive model capable of representing genotype-phenotype interactions pertaining to CAD development. An initial evaluation of the discriminative capacity of the feature space with respect to a binary classification problem (No CAD, CAD), although not complete, shows that: (i) kernel-based classification provides more accurate results as compared with neural network-based and decision tree-based modelling, and (ii) appropriate input refinement by feature ranking has the potential to increase the sensitivity of the model.