Treatment response prediction in rectal cancer patients: A radiomics study of multimodality imaging methods

The present work aims to assess the correlation of radiomics textural features derived from computed tomography (CT), magnetic resonance imaging (MRI), and endorectal ultrasound (EUS) images, combined with dosimetric and clinical features, to predict treatment response in patients with rectal cancer using machine learning algorithms.

Data from 84 individuals diagnosed with locally advanced rectal cancer (LARC) were utilized, and radiomic features were extracted from the specified region of interest. Feature selection was performed using the Least Absolute Shrinkage and Selection Operator (Lasso), Minimum Redundancy Maximum Relevance (MRMR), and Recursive Feature Elimination (RFE). Predictive modeling employed machine learning algorithms, including Support Vector Machine (SVM) and Logistic Regression (LR). Model performance was assessed based on metrics including accuracy (ACC), area under the receiver operating characteristic curve (AUC), precision, sensitivity, and specificity.

For CT images, the MRMR method (for original images) and RFE (with a wavelet filter), combined with the LR model, achieved the best performance (ACC: 0.79; AUC: 0.78). The highest predictive performance for MRI radiomic features was obtained using MRMR and the SVM model for original images (ACC: 0.88; AUC: 0.87). Furthermore, for images with the wavelet filter, the combination of RFE and the LR model yielded the best results (ACC: 0.78; AUC: 0.87). For EUS images, the MRMR and LR models showed the best predictive performance for both original (ACC: 0.89; AUC: 0.89) and filtered images (ACC: 0.81; AUC: 0.80).

The findings indicate that radiomics features obtained from pretreatment CT, MRI, and EUS images have the potential to accurately predict treatment response in patients with LARC. The SVM and LR classifiers, when combined with MRMR and RFE feature selection algorithms and the wavelet filter, demonstrated robust predictive performance. Among the different imaging modalities, EUS produced the best results in terms of ACC and AUC values.