Soutenance de thèse : Jiging Huang

Virtual liver biopsy for Chronic liver disease monitoring by using mpMRI-based radiomic

Doctorant : Jiging Huang

Laboratoire INSA : CREATIS
Ecole doctorale : ED205 EDISS

Chronic liver disease (CLD) represents a broad spectrum of diseases involving different etiologies. These diseases are characterized by histological features such as inflammation, fibrosis, steatosis, ballooning, or iron overload. Among them, inflammation plays a critical role in the early liver fibrosis process, and fibrosis affects the CLD prognosis and treatment strategy. Although liver biopsy is the gold standard for the diagnosis of CLD, Its invasiveness limits its clinical use. Therefore, an alternative noninvasive, sensitive, and specific remains an unmet medical need.
Magnetic resonance imaging (MRI), especially with diffusion-weighted imaging (DWI) appears currently as an interesting imaging technique to detect CLD-related features. The objective of this thesis is to develop the concept of virtual biopsy to grade inflammation and fibrosis in CLD. To achieve this, the thesis is divided into two parts.
Firstly, using IVIM single sequence study, we studied the standard and advanced DWI’s parameters with different fitting approaches and diffusion models, and then the relationship between CLD- related features and DWI parameters were investigated. Significant differences were found between the groups with different degrees of fibrosis. The top four significant differences parameters were selected to build classifiers to characterize fibrosis.
Secondly, from multiple MRI sequences, a radiomics approach involving extraction of several feature combinations from conventional T1w or T2w images as well as proton density fat fraction, T2* and diffusion parameter maps was investigated. The best combinations were then searched to classify inflammation and fibrosis using random forest.
This study validated the utilization of multiparametric MRI for fibrosis and inflammation severity grading and proposed two effective classifiers for them.