Deep learning reveals relationships between muscle and adipose tissue and their associations with age, body mass index, and low back pain: A cross-sectional study in 10,835 subjects.
Authors
Affiliations (6)
Affiliations (6)
- Department of Mechanical Engineering, Stanford University, Stanford, California, United States of America.
- Department of Radiology, Stanford University, Stanford, California, United States of America.
- Department of Electrical Engineering, Stanford University, Stanford, California, United States of America.
- Department of Orthopaedic Surgery, Stanford, California, United States of America.
- Department of Bioengineering Engineering, Stanford University, Stanford, California, United States of America.
- Department of Biomedical Data Science, Stanford University, Stanford, California, United States of America.
Abstract
Existing clinical imaging datasets and recent innovations in automated data analysis present a unique opportunity to analyze population-level body composition. We used a large computed tomography (CT) dataset to provide construct validity for artificial intelligence-based opportunistic assessments of body composition, report normative body composition trends, and investigate body composition differences between low back pain and matched control populations at a significantly greater scale than previous work. We applied a deep-learning model to opportunistically compute skeletal muscle, visceral adipose tissue, and subcutaneous adipose tissue cross-sectional areas (CSA) in 23,508 subjects who underwent routine clinical abdominal CT scans and analyzed body composition trends with age and body mass index (BMI). We identified a low back pain cohort using patient electronic health records and created an age-, sex-, and BMI-matched control cohort. We tested for differences in muscle CSA, ratio of visceral to subcutaneous adipose tissue (VAT/SAT), and ratio of muscle to total abdominal adipose tissue (Muscle/TAT). We identified 10,835 (n = 6,121 female) subjects for normative body composition analysis, 2,622 (n = 1,623 female) subjects with low back pain diagnosis, and 2,577 (n = 1,594 female) matched controls. Our population reproduced established rates (5-10% per decade) of muscle loss with age and increasing VAT/SAT ratio in older adults. Our low back pain cohort, 25 times larger than most prior imaging-based low back pain studies, had significantly lower (p < 0.0001) Muscle/TAT ratios than matched controls, albeit with a low effect size (η2 = 0.004). The low back pain and control groups did not differ in height-squared normalized muscle CSA or VAT/SAT ratio. We showed multiple body composition relationships that matched accepted trends, demonstrating the validity of opportunistic analysis of clinical data for large-scale investigations. In the largest low back pain study of its kind, no meaningful average difference in bulk muscle quantity was detected between low back pain and control populations.