Treatment response prediction using MRI‐based pre‐, post‐, and delta‐radiomic features and machine learning algorithms in colorectal cancer

Objectives We evaluate the feasibility of treatment response prediction using MRI-based pre-, post-, and delta-radiomic features for locally advanced rectal cancer (LARC) patients treated by neoadjuvant chemoradiation therapy (nCRT). Materials Methods This retrospective study included 53 LARC divided into a training set (Center#1, n = 36) external validation (Center#2, 17). T2-weighted (T2W) MRI was acquired all patients, 2 weeks before 4 after nCRT. Ninety-six radiomic features, including intensity, morphological second- high-order texture were extracted from segmented 3D volumes T2W MRI. All harmonized ComBat algorithm. Max-Relevance-Min-Redundancy (MRMR) algorithm used as feature selector k-nearest neighbors (KNN), Naïve Bayes (NB), Random forests (RF), eXtreme Gradient Boosting (XGB) algorithms classifiers. The evaluation performed area under receiver operator characteristic (ROC) curve (AUC), sensitivity, specificity accuracy. Results In univariate analysis, highest AUC in 0.78, 0.70, 0.71, GLCM_IMC1, shape (surface volume) GLSZM_GLNU respectively. multivariate RF KNN achieved (0.85 ± 0.04 0.81 0.14, respectively) among pre- post-treatment features. delta-radiomic-based model (0.96 0.01) followed NB 0.04). Overall. Delta-radiomics model, outperformed both (P-value <0.05). Conclusion Multivariate analysis machine learning could potentially be undergoing also observed that classifiers can powerful biomarkers LARC.