Automated Radiomics Based Clinically Significant Prostate Cancer (csPCa) Grade Classification from Biparametric MRI

Prostate cancer (PCa) is one of the most common cancers in males, and clinically significant PCa (csPCa) is one of the leading causes of death worldwide. Multi-parametric MRI (mp-MRI) has gained widespread acceptance as the primary diagnostic method for prostate cancer diagnosis. However, there are still difficulties with the widespread utilization of contrast agents based on gadolinium that brings on nephrogenic systemic fibrosis, increased complexity of interpretation, and lower productivity. To overcome the drawbacks of mp-MRI, Bi-parametric MRI (bp-MRI) is an emerging solution. Here, we propose a fully automated csPCa grade classification framework that utilizes bp-MRI for segmenting csPCa lesions and uses radiomics features. We use 124 different radiomics features from 8 feature categories and select statistically significant features to train three machine learning classifiers namely SVM, KNN, Bayesian, XGBoost and Random forest. Our framework with XGBoost classifier achieved a classification accuracy of 96% in distinguishing clinically significant PCa (ISUP 4+5) and non-significant PCa (ISUP 2+3) using the predicted maps drawn by our segmentation method.