2681 - Machine Learning Model Based on Whole Brain Radioms to Identify High Risk of Short-Term Intracranial Progression after SBRT for Brain Metastases
Presenter(s)
Q. Wang1, S. Huang2, Y. Sun3, G. Wang1, L. Zhao3, Z. Yuan3, Y. Song3, and M. Yan3; 1National Clinical Research Center for Cancer, Tianjin’s Clinical Research Center for Cancer, Tianjin Medical University Cancer Institute & Hospital, Tianjin, China, 2Tianjin Medical University Cancer Institute & Hospital, Tianjin, China, 3Department of Radiation Oncology, Tianjin Medical University Cancer Institute & Hospital, National Clinical Research Center for Cancer, Key Laboratory of Cancer Prevention & Therapy, Tianjin's Clinical Research Center for Cancer, Tianjin, China
Purpose/Objective(s): This study aimed to construct and validate a nomogram prediction model to identify patients at high risk of short-term intracranial progression (distant brain failure, DBF) following stereotactic radiotherapy for brain metastases using integrating whole-brain radiomics derived from pre-radiotherapy MRI with clinical characteristics.
Materials/Methods: A retrospective analysis was conducted on clinical data, radiotherapy plans, and pre-radiotherapy MRI images of 156 patients with brain metastases. Patients were classified into two groups based on the presence or absence of new intracranial metastatic lesions in six months. These were then randomly divided into a training set and a testing set in an 8:2 ratio. A total of 2,017 radiomic features were extracted from the whole brain tissue excluding the tumor region (Brain tissue of whole brain substrating PGTV, BTWB-PGTV) from preradiotherapy MRI images. Clinical data are subjected to univariate and multivariate analysis based on logistic regression. Finally, a nomogram model was developed by integrating the clinical features model with the radiomics model. The models were evaluated using the ROC curve and Clinical Decision Curve Analysis (DCA).
Results: Among the 156 patients included in the study, 32 experienced short-term DBF, while 124 did not. In the training set, there were 98 patients (79.04%) without short-term DBF and 26 patients (20.96%) with short-term DBF. In the testing set, 26 patients (81.25%) did not have short-term DBF, and 6 patients (18.75%) did. A total of 2017 radiomic features were extracted from the BTWB-PGTV of each patient. After Pearson correlation coefficient screening in the training set data, 392 radiomic features were retained. Further LASSO regression is applied to select the final features utilized for constructing the radiomics model. Multivariate analysis of clinical features showed that age, KPS score, and the maximum volume of a single lesion were the main factors affecting DBF (p < 0.05), therefore, these three characteristics were used in the construction of the clinical feature model. A combined nomogram model was developed integrating both BTWB-PGTV radiomics and clinical features models. All three models demonstrated robust predictive capabilities, with the combined nomogram model exhibiting superior performance. The AUC of the ROC curve was 0.906 for the training set and 0.904 for the testing set. Clinical Decision Curve Analysis revealed that the predictive outcomes of the models provided positive clinical utility, while the calibration curves indicated good agreement across the models.
Conclusion: The radiomics model derived from BTWB-PGTV of pre-radiotherapy MRI possesses the potential to predict short-term DBF in patients with brain metastases. Combining with the clinical features, the nomogram model demonstrates enhanced performance. Multicenter prospective clinical trials (ChiCTR2500096898) are currently enrolled to accuracy and generalizability of the model.