Electronic Theses and Dissertations

Date

2024

Document Type

Thesis

Degree Name

Master of Science

Department

Biomedical Engineering

Committee Chair

Aaryani Tipirneni-Sajja

Committee Member

John L Williams

Committee Member

Carl Herickhoff

Committee Member

Deepak Venugopal

Abstract

Cirrhosis is the final stage of liver disease requiring liver transplantation as treatment. In the past, models based on clinical laboratory values would determine the eligibility of liver transplant candidates. However, several patients can still suffer from liver transplantation complications leading to death. Sarcopenia, the loss of muscle mass, and frailty, the loss of muscle function, are common cirrhosis complications, but recently muscle area estimated at the third lumbar level has been studied as a prognostic to predict liver transplant mortality. Muscle fatty infiltration has been hypothesized to be a descriptor of muscle quality. However, a major bottleneck for implementing studies investigating the links among liver disease, muscle area, fatty infiltration, and frailty is the time-consuming process of muscle segmentation. Hence, the aim of this study is to develop an automatic, deep learning algorithm capable of segmenting MRI images of different contrasts for the determination of muscle mass and fatty infiltration. A secondary aim is to determine the relationship between muscle area and fatty infiltration with frailty and sex. Transfer learning and incremental learning techniques were compared to determine which method would best segment two different types of MRI images of visually opposite contrast simultaneously. Three models’ design and training parameters were optimized with Bayesian optimization and bandit methods. Two models were trained on two opposite contrast image sets independently, while one dual-contrast model was developed. The dual-contrast model had comparable muscle mass estimation to individual models, with slightly higher variability. There were statistical differences in muscle mass between frail and non-frail groups. When including sex, there was a difference between frail males and non-frail males as well as frail females and non-frail males. No difference in fatty infiltration was found among frailty groups and sex. In conclusion, an automated model was developed capable of segmenting paraspinal muscles on MRI images of different contrasts, and there were significant differences in muscle area among frailty groups.

Comments

Data is provided by the student.

Library Comment

Dissertation or thesis originally submitted to ProQuest.

Notes

Embargoed unitl 09-23-2025

Download

Available for download on Tuesday, September 23, 2025

Share

COinS