Integrating topological features to enhance cardiac disease diagnosis from 3D CMR images

Abstract

[en] Persistent homology is a technique from the field of algebraic topology for the analysis and characterization of the shape and structure of datasets in multiple dimensions. Its use is based on the identification and quantification of topological patterns in the dataset across various scales. In this thesis, persistent homology is applied with the objective of extracting topological descriptors from three-dimensional cardiovascular magnetic resonance (CMR) imaging. Thereafter, topological descriptors are used for the detection of cardiovascular diseases by means of Machine Learning (ML) techniques. Radiomics has been one of the recently proposed approaches for disease diagnosis. This method involves the extraction and subsequent analysis of a significant number of quantitative descriptors from medical images. These descriptors offer a characterization of the spatial distribution, texture, and intensity of the structures present in the images. This study demonstrates that radiomics and topological descriptors achieve comparable results, providing complementary insights into the underlying structures and characteristics of anatomical tissues. Moreover, the combination of these two methods leads to a further improvement of the performance of ML models, thereby enhancing medical diagnosis.