The Role of Artificial Intelligence in Magnetic Resonance Imaging of Chronic Liver Disease
Machine learning (ML) and especially deep learning (DL) techniques such as Convolutional Neural Networks (CNNs) have emerged as an auspicious tool in medical imaging due to their unprecedented capabilities in pattern recognition and thus the extraction of predictive imaging features from MRI data.
The role of Gadoxetic-enhanced MRI imaging in Chronic liver disease (CLDs) is lately gaining relevance since it has become a powerful tool not only for providing functional information about the liver parenchyma, but in addition prediction of transplant-free-survival-time, transplant-survival-time and possibly time to hepatic decompensation.
The aim of this PhD thesis is to evaluate the added value of Deep Learning (DL) algorithms in the early diagnosis of CLD, and determine if it helps to predict complications such as decompensation and liver-related mortality. Furthermore, we would expect that information generated by DL will help us to understand, how CLD develops and which risk factors predispose to the disease. Therefore, the interdisciplinary aspect of linking results of DL methods with biological understanding and potentially novel hypotheses regarding underlying mechanisms is central to this PhD thesis. To this end, the thesis work is embedded in multidisciplinary research.
The thesis research has three primary objectives:
Objective 1: Analyze and evaluate a multitask learning framework (a form of CNN) for automatic segmentation of liver vessels on portal venous T1-weighted MRI sequences and unenhanced MRI T1 sequences. Evaluate the prediction of both contrast enhanced images and vessel segmentations from unenhanced MRI T1 sequence, assessing utility and role in clinical diagnosis of liver disease.
Objective 2: Analyze the association of liver vessel volume and its changes to clinical parameters in chronic liver disease and liver related events such as decompensation and transplant-free survival.
Objective 3: Analyze the role of imaging characteristics of the spleen as additional predictors in advanced chronic liver disease, with a focus on spleen texture, and the predictive value for clinical data including labor parameters and FLIS.
Methods and Skills:
clinical studies; liver MRI; artificial intelligence; deep learning
Poetter-Lang S, Bastati N, Messner A, Kristic A, Herold A, Hodge JC, Ba-Ssalamah A. Quantification of liver function using gadoxetic acid-enhanced MRI. Abdom Radiol (NY) 45: 3532-3544, 2020