Pectoralis muscle deep learning on low-dose chest CT detects type 2 diabetes

A hybrid deep learning model of pectoralis muscle on low-dose chest CT detected type 2 diabetes with an AUC of 0.746 in external validation, significantly outperforming a conventional metric (AUC 0.515).

• Multi-center study (n=1,209) with separate training, internal validation, and external validation cohorts.
• The hybrid model's AUC was 0.746 vs. 0.515 for baseline muscle metrics (p<0.001); no significant difference vs. DL-only model.
• SHAP analysis showed lower muscle CT attenuation and deep learning signatures were independent risk indicators.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.