Radiomics analysis for prostate cancer classification in multiparametric magnetic resonance images

The objective of the study is to evaluate local binary pattern (LBP) features on T2-weighted magnetic resonance (MR) images in differentiating prostate cancer (PCa) from benign prostatic tissues, and to compare its performace with commonly used radiomics features. A total of 63 patients were studied who undergone multiparameter magnetic resonance scans, including T2-weighted and diffusion-weighted imaging (DWI). Two-dimensional regions of interest (ROIs) of lesions identified by proprective clinical readings were manually outlined by a radiologist. Radiomics features were extracted from ROIs on T2-weighted images and apparent diffusion coefficient (ADC) maps derived from DWI images, respectively. Feature selection was conducted to remove irrelevant or redundant information. Selected features were used to train classifiers. Thorough experiments were conducted to find the best combination of feature selection methods and classifiers. LBP features were extracted from ROIs on T2-weighted images and support vector machine (SVM) was used for the classification. Cross-validation and receiver operating characteristic (ROC) analysis were used to evaluate the performance of the classification. The area under the ROC curve (AUC) of the best radiomics features and LBP features were 0.92 and 0.94, respectively. The study demonstrated that the LBP features were effective for the classification of PCa and benign prostatic tissues. Further studies are warranted to investigate the value of the LBP features in PCa assessment.