Cardiac motion analisys from TAG-MRI using radiomics

Abstract

Abnormalities in heart wall motion are often related to cardiovascular disease (CVD), the first cause of death worldwide. In this context, tag-MRI imaging technique has the potential to become the gold-standard for quantifying regional function and therefore to enable reliable stratification of CVDs patients. However, there is a lack of comprehensive image analysis methods that can analyse the cardiac motion from tag-MRI. On the other hand, the radiomics paradigm has recently shown great promise for patient stratification in the presence of complex diseases in cardiac MRI. In this project, the feasibility of using radiomics and machine learning for cardiac stratification in tag-MRI is investigated through a number of methods and experiments. First, a learning radiomics-based approach is implemented to predict the motion landmarks conventionally defined through semi-automatic methods, indicating limited correlations between the two types of variables. Subsequently, the potential of radiomics for cardiac motion stratification in tag-MRI is implemented based on feature selection and principal component analysis. The results obtained based on a public database of 15 tag-MRI cases show that, unlike the motion landmarks used in previous research, radiomics features estimated from tag-MRI have the potential for discriminating between distinct cardiovascular subgroups. This thesis represents the first proof-of-concept study for deeper phenotyping and advanced stratification of cardiac motion using tag-MRI. Future work includes more extensive validation with larger clinical samples and diverse CVD subgroups