Prediction of pathological complete response to neoadjuvant chemoimmunotherapy in non-small cell lung cancer using

backgroundPredicting the response to neoadjuvant chemoimmunotherapy in patients with resectable non-small cell lung cancer (NSCLC) facilitates clinical treatment decisions. Our study aimed to establish a machine learning model that accurately predicts the pathological complete response (pCR) using

methodsWe retrospectively included 210 patients with NSCLC who completed neoadjuvant chemoimmunotherapy and subsequently underwent surgery with pathological results, categorising them into a training set of 147 patients and a test set of 63 patients. Radiomic features were extracted from the primary tumour and lymph nodes. Using 10-fold cross-validation with the least absolute shrinkage and selection operator method, we identified the most impactful radiomic features. The clinical features were screened using univariate and multivariate analyses. Machine learning models were developed using the random forest method, leading to the establishment of one clinical feature model, one primary tumour radiomics model, and two fusion radiomics models. The performance of these models was evaluated based on the area under the curve (AUC).

resultsIn the training set, the three radiomic models showed comparable AUC values, ranging from 0.901 to 0.925. The clinical model underperformed, with an AUC of 0.677. In the test set, the Fusion_LN1LN2 model achieved the highest AUC (0.823), closely followed by the Fusion_Lnall model with an AUC of 0.729. The primary tumour model achieved a moderate AUC of 0.666, whereas the clinical model had the lowest AUC at 0.631. Additionally, the Fusion_LN1LN2 model demonstrated positive net reclassification improvement and integrated discrimination improvement values compared with the other models, and we employed the SHapley Additive exPlanations methodology to interpret the results of our optimal model.

conclusionsOur fusion radiomics model, based on