Patients with multiple myeloma commonly undergo prolonged, multi-line and multidrug therapies that can adversely affect physiological reserve, physical function, and overall health. Routine whole-body magnetic resonance imaging (WBMRI) performed for disease assessment captures nondiseased organs and tissues that can be opportunistically analyzed to derive quantitative measures of body composition. Recent advances in WBMRI acquisition and standardization, alongside improvements in artificial intelligence (AI) for automated segmentation, create an opportunity to extract scalable imaging biomarkers from existing clinical scans.
The authors report the development of an AI-based deep-learning pipeline designed to produce automated, quantitative image-derived phenotypes from nondiseased tissues on routine clinical WBMRI in patients with myeloma. The pipeline performs automated organ and tissue segmentation in three dimensions, enabling measurement of body composition compartments such as abdominal skeletal muscle, abdominal subcutaneous adipose tissue, and visceral adipose tissue. The publication states this to be the first reported application of an automated pipeline to derive these body composition metrics from WBMRI in myeloma patients.
When applied longitudinally across treatment, the pipeline detected statistically significant temporal changes in body composition (P < .001). The dominant pattern described was a decrease in abdominal skeletal muscle (ASM) over time, accompanied by transient increases in both abdominal subcutaneous adipose tissue and visceral adipose tissue (VAT). These longitudinal measurements demonstrate that routinely acquired diagnostic imaging can track dynamic shifts in muscle and fat compartments during the course of myeloma therapy.
The study examined relationships between baseline and longitudinal body composition metrics and clinical outcomes. Greater baseline reserves of ASM were associated with better progression-free survival (hazard ratio [HR], 0.60; 95% confidence interval [CI], 0.40–0.89). Similarly, higher baseline abdominal subcutaneous adipose tissue correlated with improved PFS (HR, 0.67; 95% CI, 0.46–0.98). In contrast, increases in visceral adipose tissue over time were linked to inferior PFS (HR, 2.89; 95% CI, 1.65–5.09). These effect estimates indicate differential prognostic associations for muscle and adipose compartments measured opportunistically from WBMRI.
The authors report that opportunistic body composition phenotyping from diagnostic WBMRI functioned as a scalable biomarker for risk stratification, with a concordance index of 0.725. This suggests the derived imaging metrics contributed meaningful prognostic information for progression-free survival beyond standard assessments, supporting use in mechanistic studies and potential clinical risk models.
This observational study is registered at ClinicalTrials.gov under identifier NCT02403102. Conflict-of-interest disclosure notes that two authors (N.B. and B.W.) are cofounders of Orcino Health Ltd.; the Royal Marsden Hospital has no financial relationship with Orcino Health Ltd. The remaining authors declared no competing financial interests.
The findings support the feasibility of extracting quantitative body composition metrics from routine WBMRI using an AI pipeline and demonstrate clinically relevant associations between those metrics and progression-free survival in myeloma. Specifically, greater abdominal skeletal muscle and higher abdominal subcutaneous adipose tissue at baseline were associated with improved PFS, whereas increasing visceral adipose tissue during treatment associated with worse PFS. The authors present the approach as scalable and suitable for mechanistic research and risk stratification.
Limitations and specific methodological details (including cohort size, inclusion criteria, imaging protocols, model training and validation procedures, and adjustment variables in outcome models) were not fully reported in the abstract. For complete information on cohort characteristics, statistical methods, and supplementary analyses, consult the full text of the article.
An AI-driven deep-learning pipeline applied to routine WBMRI in patients with multiple myeloma can generate automated, quantitative body composition metrics. Longitudinal application revealed treatment-associated declines in abdominal skeletal muscle and transient fat increases. Baseline muscle and subcutaneous fat reserves were associated with improved progression-free survival, while increases in visceral fat over time predicted worse outcomes. The authors propose opportunistic WBMRI phenotyping as a scalable imaging biomarker for risk stratification and mechanistic investigation in myeloma care.
(Trial registration: NCT02403102.)