Correlation of whole-body fat content and its distribution with lumbar fat content under different age, gender and BMI conditions: quantitative analysis of MRI based on deep learning.
Authors
Affiliations (9)
Affiliations (9)
- Department of Radiology, Shanghai Tenth People's Hospital, School of Medicine, Tongji University, No.301 Yangchang Road, Shanghai, 200072, China.
- Department of Radiology, Ningbo No.2 Hospital, No. 41 Northwest Street, Haishu District, Ningbo, Zhejiang, 315010, China.
- Ningbo Key Laboratory of Digital Imaging and Medical-Engineering Interdisciplinarity, Ningbo, Zhejiang, 315010, China.
- Department of Radiotherapy Technology, Ningbo No.2 Hospital, Ningbo, Zhejiang, 315010, China.
- Department of Research and Development, Shanghai United Imaging Intelligence Co., Ltd, Shanghai, 200232, China.
- Paul C. Lauterbur Research Center for Biomedical Imaging, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, 518055, China.
- State Key Laboratory of Biomedical lmaging Science and System, Shenzhen, 518055, China.
- Department of Radiology, Ningbo No.2 Hospital, No. 41 Northwest Street, Haishu District, Ningbo, Zhejiang, 315010, China. [email protected].
- Department of Radiology, Shanghai Tenth People's Hospital, School of Medicine, Tongji University, No.301 Yangchang Road, Shanghai, 200072, China. [email protected].
Abstract
The correlation between abnormal distribution of systemic fat deposition and fat content in the vertebral body remains unclear, and AI segmentation offers novel approach for large-scale quantitative analysis. To explore the effects of gender, age, and body mass index on the correlation between whole-body fat content/distribution and lumbar spine fat content using deep learning based MRI quantification technology. One hundred thirteen volunteers (63 males/50 females; 47.7 ± 13.3 years) who underwent coronal whole-body and lumbar sagittal MRI examination were enrolled. Through automatic segmentation and quantification, total adipose tissue (TAT), body-fat percentage (BF%), liver fat fraction (LFF), subcutaneous fat fraction (SAT/TAT), internal fat fraction (IAT/TAT), abdominal subcutaneous/visceral fat ratio (ASVFR), erector spinae-multifidus fat fraction (ESFF), psoas major fat fraction (PMFF) and lumbar vertebral fat fraction (VFF) were obtained. Pearson correlation was then used for stratified analysis based on sex, age (≤ 45, 46-59, ≥ 60 years) and BMI (underweight/normal/over-weight/obese). All preset variables were entered into a Lasso model (10-fold cross-validation) to identify independent VFF determinants and to calculate R² and the root-mean-square error (RMSE). VFF correlated moderately with ESFF and PMFF (r_<sub>ESFF</sub>=0.53, r_<sub>PMFF</sub>=0.39), weakly with TAT and BF% (r_<sub>TAT</sub>=0.23, r_<sub>BF%</sub>=0.26), and rarely with LFF (r = 0.08). Fat distribution exhibits "site-specificity", with a relatively significant correlation between ASVFR and VFF (r=-0.40), a weak positive correlation between IAT/TAT and VFF (r = 0.29), and a slightly negative association between SAT/TAT and VFF (r=-0.29). Correlations between VFF and relevant indicators exhibited stratified differences by gender, age, and BMI, with significant gender-specific variations, insignificant age-specific variations, and ESFF being the sole indicator prominently and positively correlated with VFF across all BMI strata. The multivariate model (R² = 0.63) identified ESFF, PMFF, and ASVFR as the core drivers of VFF, rather than whole-body fat or liver fat. Paravertebral muscle fat infiltration and visceral-fat predominance are associated with higher lumbar marrow fat accumulation, which tends to be more effective in the female, elderly and overweight groups. The constructed multi-variable model may provide a quantitative imaging framework for future studies investigating vertebral marrow fat and bone health.