Improvement of machine learning models for predicting high-grade subtypes of lung adenocarcinoma based on delta...

To evaluate the effectiveness of delta radiomics in predicting high-grade components in lung adenocarcinoma and to develop a robust machine learning model for clinical application.

This retrospective multi-center cohort study included lung cancer patients from three hospitals who had pre-surgery CT follow-up scans. Training (n = 491) and validation (n = 210) were performed using cases from Center 1, and testing was conducted using cases from Centers 2 and 3 (n = 92). Radiomic features were extracted from baseline and follow-up CT images, and delta radiomic features were calculated. The LASSO algorithm was used for radiomic feature selection, and rad-score and delta rad-score were constructed. Significant clinical and radiomic features were combined to build the final machine learning model. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), DeLong test, decision curve analysis (DCA), and integrated discrimination improvement (IDI) analysis.

In the external test cohort, the integrated machine learning model constructed based on clinical features (CTR, smoking status, maximum diameter of the solid component), rad-score, and delta rad-score showed that the random forest model performed the best, with an AUC of 0.91. The random forest model outperformed the clinical model (AUC = 0.80), rad-score (AUC = 0.79), and delta rad-score (AUC = 0.81). DCA and IDI indicated that the random forest model provides superior clinical benefit and improvement.

Delta radiomics significantly aids in identifying high-grade subtypes of lung adenocarcinoma. The integrated machine learning model offers an effective approach for prediction of high-grade components, with potential clinical implications.

This study presents a novel application of delta radiomics to predict high-grade lung adenocarcinoma, which may influence surgical management and improve patient outcomes.