689483 event 1775503861 1775503901 Khushi Shah
BME MS Thesis Defense Presentation
Date: 2026-04-14
Time: 9:00AM - 10:30AM
Location / Meeting Link: Conference Room N657 HSRB II, Emory University, Atlanta, GA Zoom Link: https://emory.zoom.us/j/94212657259

Committee Members:
Dr. Anant Madabhushi (advisor); Dr. Laxmi Prasad Dasi; Dr. Hanjoong Jo; Dr. Puja Kiran Mehta


Title: Multi-Organ Fat quantification from Whole-Body MRI for longitudinal prediction of Major Adverse Cardiovascular Events (MACE) 

Abstract:
Introduction: Multi-organ fat quantification can improve longitudinal cardiovascular risk stratification by capturing tissue-level metabolic heterogeneity beyond what conventional adiposity measures and clinical risk scores provide. In this work, we propose a population-scale radiomic framework to understand inter-organ fat dynamics using whole-body MRI for cardiovascular diseases. Methods: We integrate automated deep learning organ segmentation to extract voxel-wise signals across five organs- heart, liver, pancreas, kidneys, and intestine. Fat fraction maps were partitioned into population-anchored density-stratified subregions, from which morphological and intensity radiomic features were derived per subject and reduced via LASSO bootstrap stability selection. Organ-specific and multi-organ Cox Proportional Hazards models were trained on a geographically stratified split and evaluated for 10-year incident MACE prediction, benchmarked against conventional adiposity measures and established clinical risk factors. Results: Combined multi-organ risk score yielded the highest discrimination for 10-year MACE (C-index = 0.636; HR = 1.76, p < 0.001), exceeding the performance of individual organ models (heart C-index= 0.635, pancreas C-index= 0.605, liver C-index= 0.598, kidney C-index= 0.590, intestine C-index= 0.585) and consistently outperforming visceral adipose tissue alone (C-index= 0.602) and subcutaneous fat alone (C-index= 0.50) model. Integrated model (inter-organ fat + clinical model), improved discrimination to C-index= 0.772 (ΔC-index= +0.7%), over the clinical alone model. The dominant selected features across organs were morphological and the fat subregion volume. These results establish that shape-based characterization of focal high-density ectopic fat deposits, rather than scalar fat fraction, is the primary imaging determinant of cardiovascular risk, and motivate integration of multi-organ fat radiomics into population imaging pipelines for personalized preventive cardiology. 

]]> Multi-Organ Fat quantification from Whole-Body MRI for longitudinal prediction of Major Adverse Cardiovascular Events (MACE) 

]]> 221981 1788 111531