Incremental diagnostic value of multiregional single-slice CT muscle areas over L3 for sarcopenia: a deep learning-based segmentation study

Abstract

Objective To compare the diagnostic performance of multiregional CT-based muscle assessment with conventional single-level (L3) evaluation. Materials and methods This retrospective study included 83 adults who underwent multiregional non-contrast CT and completed sarcopenia assessments based on the Asian Working Group for Sarcopenia (AWGS) 2019 criteria. Skeletal muscle areas at six anatomical levels (T4, L3, ASIS, femoral head, midthigh, and proximal calf) were automatically quantified using a deep learning-based segmentation software (DeepCatch). The diagnostic performance of single-slice muscle areas was evaluated using the area under the receiver operating characteristic curve (AUC). Pearson's or Spearman's correlation coefficients were analyzed to assess the relationship between CT-derived muscle metrics and functional status. Results Multiregional muscle assessment demonstrated the highest diagnostic performance. In a clinical prediction model, models incorporating three-site and six-site muscle areas improved discrimination compared with the clinical base model (Delta AUC 0.123 and 0.136, respectively), and these improvements remained significant after BH-FDR adjustment (both q = 0.034). In contrast, the addition of the midthigh muscle area showed a modest improvement (Delta AUC 0.089; p = 0.029), which did not remain significant after FDR adjustment (q = 0.064), and L3 muscle area provided limited incremental value. Conclusion Multiregional CT-based muscle assessment provides improved diagnostic performance for sarcopenia compared with single-level evaluation. Lower-extremity muscle measurements, particularly at the midthigh, contribute to this improvement, whereas reliance on L3 alone may be insufficient.