P352 - LEVERAGING ROUTINE MRI TO PREDICT BODY COMPOSITION IN BREAST CANCER: A NUTRITIONALLY-INFORMED IMAGING APPROACH

P352

LEVERAGING ROUTINE MRI TO PREDICT BODY COMPOSITION IN BREAST CANCER: A NUTRITIONALLY-INFORMED IMAGING APPROACH

M. M. M. De Giulio^1,2,*, C. Serantoni^1,2, A. Riente^1,2, D. Hatem^1,2, T. Marchetti², S. Magno², A. Filippone², C. Rossi ², M. Rossi ², A. Di Leone², G. Franceshini², A. Franco², V. Castagnetta², P. Belli ², M. Conti ², M. De Spirito^1,2, G. Maulucci ^1,2 on behalf of Metabolic Intelligence Lab, Rome

¹Univerità Cattolica del Sacro Cuore, ²Fondazione Policlinico "A. Gemelli", Rome, Italy

Rationale: Monitoring body composition is essential in managing breast cancer patients, especially for early detection of treatment-induced sarcopenia. While BMI and waist-to-hip ratio (WHR) provide general estimates, MRI allows localized assessment of tissue compartments. This study aims to develop a predictive model integrating anthropometric data with MRI-derived features to estimate body composition indices, maximizing the value of imaging routinely acquired during diagnosis and follow-up.

Methods: Fifty breast cancer patients undergoing standard imaging were enrolled. For each, a single axial slice at the level between the manubrium and the sternal body was selected. The intermammary fat region was segmented from T2-weighted images, and a region of interest on the pectoralis muscle was extracted from fat-suppressed T1-weighted sequences. Quantitative features describing signal intensity, distribution, and geometry were computed and combined with BMI and WHR. Principal Component Analysis (PCA) was used for dimensionality reduction. Regression models were trained to predict the FFM%/FAT% ratio and Fat-Free Mass Index (FFMI = FFM%/height²).

Results: The integration of MRI features with anthropometric data yielded robust prediction performance. The model achieved an R² of 0.82 and MAE of 4.2% for the FFM%/FAT% ratio, and R² of 0.79 with MAE of 0.41 kg/m² for FFMI. Using a single MRI slice ensured clinical efficiency and reproducibility without increasing patient imaging burden.

Conclusion: This study proposes a non-invasive method to assess body composition using routine MRI and anthropometry. By detecting early lean mass loss, the model supports personalized, nutrition-aware management in breast cancer care without additional diagnostic procedures.

Disclosure of Interest: None declared