Automated segmentation reveals sex-divergent age-related decline in temporalis muscle volume among Asian adults

Temporalis muscle measurements are recognized biomarkers of sarcopenia. However, manual assessments on computed tomography (CT) are time-consuming and prone to interobserver variability, and population-specific normative data for Asian populations are limited. A Multi-Scale Residual Dense Attention U-Net was developed for automated temporalis muscle segmentation from head CT images and validated using fivefold cross-validation. Age- and sex-specific normative values were derived from 320 Taiwanese adults (160 males and 160 females) spanning eight age groups from 20 to 99 years. The model achieved a Dice coefficient of 0.914 and an F1 score of 0.958. In males, peak temporalis muscle volume (TMV) was 36.9 mL at 20-29 years, followed by a 38% decline to the oldest age group. In females, peak TMV was 27.9 mL at 30-39 years, with a 60% decline across the lifespan. Females exhibited a markedly faster rate of decline than males during midlife, with rates converging after 65 years of age. Temporalis muscle area showed a strong correlation with TMV (r = 0.882, p < 0.001). This study establishes the first comprehensive age- and sex-specific normative database for TMV in Taiwanese adults. The proposed approach enables standardized muscle mass assessment from routine head CT scans.