Explainable multimodal fusion model predicts high-grade lung adenocarcinoma patterns on CT

A multimodal fusion model integrating clinical and quantitative CT features achieved AUC 0.848 (sensitivity 84.7%, specificity 71.1%) for preoperative prediction of high-grade histologic patterns in invasive lung adenocarcinoma, with external validation AUC 0.815.

• The fusion model outperformed clinical-only and radiomics-only models, as well as standard classifiers (XGBoost, logistic regression, multilayer perceptrons).
• Decision curve analysis showed higher net clinical benefit than treat-all or treat-none strategies across threshold probabilities.

Automated summary

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