Radiogenomics Framework for Associating Medical Image Features with Tumour Genetic Characteristics

Abstract

Significant progress has been made in the understanding of human cancers at the molecular genetics level and it is providing new insights into their underlying pathophysiology. This progress has enabled the subclassification of the disease and the development of targeted therapies that address specific biological pathways. However, obtaining genetic information remains invasive and costly. Medical imaging is a non-invasive technique that captures important visual characteristics (i.e. image features) of abnormalities and plays an important role in routine clinical practice. Advancements in computerised medical image analysis have enabled quantitative approaches to extract image features that can reflect tumour genetic characteristics, leading to the emergence of ‘radiogenomics’. Radiogenomics investigates the relationships between medical imaging features and tumour molecular characteristics, and enables the derivation of imaging surrogates (radiogenomics features) to genetic biomarkers that can provide alternative approaches to non-invasive and accurate cancer diagnosis.

This thesis presents a new framework that combines several novel methods for radiogenomics analysis that associates medical image features with tumour genetic characteristics, with the main objectives being: i) a comprehensive characterisation of tumour image features that reflect underlying genetic information; ii) a method that identifies radiogenomics features encoding common pathophysiological information across different diseases, overcoming the dependence on large annotated datasets; and iii) a method that quantifies radiogenomics features from multi-modal imaging data and accounts for unique information encoded in tumour heterogeneity sub-regions. The present radiogenomics methods advance radiogenomics analysis and contribute to improving research in computerised medical image analysis.