Development and validation of multi-sequence magnetic resonance imaging radiomics models for predicting tumor response to radiotherapy in locally advanced non-small cell lung cancer.

Background: The significant heterogeneity of locally advanced non-small cell lung cancer (LA-NSCLC) poses challenges for clinical decision-making. Although various predictive methods currently exist, there remains a lack of an accurate and comprehensive approach effectively assessing the response of LA-NSCLC patients to radiotherapy. Therefore, the objective of this study was to develop a model based on multisequence magnetic resonance imaging (MRI) radiomics features to predict tumor response following radiotherapy in patients with LA-NSCLC and to evaluate its clinical utility. Methods: Data was retrospectively collected from stage III non-small cell lung cancer (NSCLC) who underwent radiotherapy and had MRI scans performed prior to treatment from November 2018 to April 2024. The patients were randomly divided into a training set and a testing set in a ratio of 7:3. Tumor response, assessed using Response Evaluation Criteria in Solid Tumors (RECIST) 1.1, classified complete or partial responses as regression sensitive (RS) and stable or progressive disease as regression resistant (RR). A total of 3,045 radiomic features were extracted and integrated from T1-weighted imaging (T1WI), T2-weighted imaging (T2WI), and diffusion-weighted imaging (DWI). Feature selection was conducted using the maximum relevance minimum redundancy (mRMR) and the least absolute shrinkage and selection operator (LASSO). Four machine learning algorithms, including logistic regression (LR), support vector machine (SVM), k-nearest neighbor (KNN), and random forest, were utilized to construct radiomic models for individual and combined sequences, with the optimal model selected. The models were evaluated with receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). Results: A total of 86 patients were included in this study, with 60 patients (RS: n=32, RR: n=28) in the training set and 26 patients (RS: n=14, RR: n=12) in the testing set. Among them, there were 45 males in the training set and 16 males in the testing set. Four, eight, six, and nine features were selected from the T1, T2, DWI sequences, and combined sequences, respectively. Among the four machine learning algorithms, the LR model demonstrated high accuracy and stability. The combined sequence model exhibited the highest predictive performance. The area under the curve (AUC) of the training set was 0.888 [95% confidence interval (CI): 0.803-0.974], and the accuracy, sensitivity, and specificity were 83.3%, 92.9% and 75%. The AUC of the testing set was 0.815 (95% CI: 0.648-0.983), and the accuracy, sensitivity, and specificity were 73.1%, 91.7% and 57.1%. Calibration curve and DCA demonstrated that the combined sequence model possessed good predictive performance and clinical utility. Conclusions: The radiomics model developed using multisequence MRI demonstrates strong potential in predicting the radiotherapy response in patients with LA-NSCLC. Notably, the combined sequence model outperforming individual sequence models in both predictive accuracy and clinical applicability.