Title: A comparison of magnetic resonance imaging and dual energy X-ray absorptiometry for the assessment of thigh muscle mass
Supervisor: Dr. Robert Boushel (and Joshua Bovard, PhD Candidate)
Second Reader: Dr. Cameron Mitchell
Abstract: Quantifying exercising muscle mass is of great interest to exercise physiologists as it facilitates relative calculations of physiological measurements and therefore comparisons accounting for different body sizes and compositions. Magnetic resonance imaging (MRI) is considered the gold standard for assessing muscle mass but manual muscle segmentation is laborious and image acquisition is expensive. Recently developed machine learning platforms can be trained with manual muscle segmentations to then automatically segment muscles with substantially reduced segmentation time. Alternatively, dual energy X-ray absorptiometry (DEXA) is a convenient and cost-effective method to assess total body and regional compositions, and DEXA thigh lean mass was highly correlated with MRI thigh muscle volume. In this major graduating paper, I aimed to compare MRI muscle cross-sectional area assessed using manual and automatic segmentations and to assess if the calculated MRI muscle volume was similar to an estimation using MRI-DEXA regression equations. Sixteen subjects underwent MRI and DEXA scans consecutively after an overnight fast. For four subjects, I calculated volumes for fourteen thigh muscles using manual segmentation, then trained the machine learning model and subsequently performed automatic segmentation for the same subjects. The calculated MRI muscle volume was compared with an estimated muscle volume derived from the DEXA thigh lean mass and a published DEXA-MRI regression equation. The muscle cross-sectional areas determined using automatic segmentation did not correlate well with those determined using manual segmentation at the proximal and distal portions of the femur, but better correlated between 20 to 80% of the femur length. The DEXA-estimated muscle volume overestimated the calculated muscle volume. These findings provide valuable insight into the amount of training required for machine learning-based automatic segmentation, as well as the validity of DEXA-MRI regression equations for estimating muscle volumes.