Ultra-low dose CT-based automated volumetric measurement of muscle, fat, and body-composition at the hip

Besides osteoporosis, body-composition contributes to fall incidences and fragility fractures, which are experienced by 40-50% women and 13-22% men. Available algorithms for muscle and subcutaneous adipose tissue (SAT) quantification are unsuitable for ultra-low dose (ULD) CT. We develop an ULD CT-based automated method using an x-ray exposure equivalent to 18 days of environmental radiation to measure gluteal muscle and SAT volumes and their ratio as body-composition at both hips. Relative thresholding, connectivity, and morphological analysis is applied on deep learning-derived likelihood maps to segment and compute percent of muscles and SAT volumes. ULD CT images of 30 participants were used for training, and images of another 30 participants, including 12 participants with matching ULD and clinical CT and five participants with repeat ULD CT scans, were used for evaluation. Reference segmentation volumes were derived from clinical CT scans using expert manual outlining. Dice scores (n=12) were 0.984 and 0.986 for muscle and SAT segmentation, respectively. Concordance correlation coefficients of 0.981, 0.996, and 0.957 were observed for muscle, SAT, and body-composition (n=12). Intraclass correlation coefficients of 0.981, 0.997, and 0.958 were observed for muscle, SAT, and body-composition in repeat scans (n=5). ULD body-composition analysis (age (mean±SD) 73.0±8.1 years; 15 females) showed that males have significantly lower (p=0.006) SAT volume (41.9±7.0%) and improved (p⪅0.001) body-composition (40.0±11.1%) compared to females (50.3±8.3% and 25.6±8.4%, respectively). Our method offers accurate measures of body-composition at the hip, and the ULD feature will facilitate its application to different research studies related to bone and metabolic diseases.