CT body composition plus machine learning predicts gallstones in dual-center study

In a dual-center cohort, an extreme trees model using CT body composition indices achieved an AUC of 0.772 (95% CI 0.714-0.822) for predicting gallstone occurrence, outperforming the best single predictor, neutrophil-to-lymphocyte ratio (NLR; AUC 0.627).

• The model integrated CT body composition indices (fat, muscle, bone at L3) with anthropometric and laboratory indices; SHAP analysis identified subcutaneous-to-total fat area ratio as the most important feature.
• The study included 1,944 patients from two centers with 3-year follow-up and external validation on a separate test cohort (n=507).
• NLR, though the best univariate predictor, had limited standalone utility (AUC 0.627).

Automated summary

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