Deep learning improves standardization and speed of CT-based body composition analysis in breast cancerArtificial intelligence helps standardize body composition analysis for breast cancer

Computed tomography (CT)-derived body composition parameters—including skeletal muscle area, muscle density, and visceral and intermuscular adipose tissue—are robust prognostic and predictive biomarkers in breast cancer. However, their clinical translation is hampered by substantial methodological heterogeneity across studies. Key sources of variability include image acquisition parameters (contrast phase, tube voltage, slice thickness), inconsistent vertebral level selection (L3 versus thoracic levels), the absence of standardized normalization methods, and fragmented diagnostic thresholds for sarcopenia, myosteatosis, and visceral obesity. This narrative review dissects these sources of variability and demonstrates how artificial intelligence (AI), particularly deep learning, provides a transformative solution. AI offers a transformative pathway towards standardization of the analytical workflow—from intelligent slice localization and multi-label tissue segmentation (including challenging compartments such as intermuscular adipose tissue) to the generation of population-specific reference curves and multimodal risk prediction. Open-source platforms such as AutoMATiCA and DAFS Express achieve human-expert accuracy with sub-second processing times, directly addressing reproducibility concerns. By shifting from fixed, population-derived cutoffs to dynamic, individualized reference systems, AI offers a clear pathway toward integrating standardized body composition analysis into routine clinical practice, ultimately advancing precision oncology for breast cancer patients.