|
|
|
 |
Search published articles |
 |
|
Showing 1 results for Machine-Learning.
E. Sadati, Ph.d., B. Hashemi, S.r. Mahdavi, A. Nikoufar, H. Abdollahi,
Volume 23, Issue 1 (1-2025)
Abstract
Background: We aimed to develop a robust prognostic model for assessing the risk of complications associated with radiotherapy in prostate cancer patients using radiomics and dosiomics feature and machine learning. Materials and Methods: A cohort of 60 patients undergoing pelvic radiation therapy was analyzed. The patients’ radiomics and dosiomics features were extracted from segmented bladder and rectum regions in CT images, as well as 3D dose distribution data, respectively. Classifier algorithms, such as eXtreme Gradient Boosting (XGBoost), Decision Tree (DT), Support Vector Machines (SVM), K-Nearest Neighbor (KNN), Logistic Regression (LR), Random Forest (RF), and Multilayer Perceptron (MLP) were used for prediction modeling. A 5-fold cross-validation method was used to evaluate the predictive classification of patients with and without proctitis and cystitis. The area under the receiver operating characteristic curve (AUC) was used for comparing models’ performance, as well assessing their specificity and accuracy metrics. Results: Various combinations of feature selection and classifier algorithms evaluated on both training and test datasets revealed that for bladder toxicity, the Relief+KNN dosiomics model, Boruta+SVM radiomics model, and the combined radiomics and dosiomics model with ANOVA+XGBoost show the highest AUCs of 0.76, 0.68, and 0.67, respectively. Regarding the rectal toxicity, the best-performing models were Boruta+KNN for dosiomics (AUC 0.83), ANOVA+RF for radiomics (AUC 0.72), and ANOVA+XGBoost for the combined radiomics and dosiomics (AUC 0.71). Conclusion: Our study demonstrated the effectiveness of diverse algorithms leveraging quantitative features extracted from CT imaging and 3D dose distribution data in predicting post-radiotherapy complications in prostate cancer patients.
|
|
