Multiclass Classification of Prostate Cancer Gleason Grades Groups Using Features of multi parametric-MRI (mp-MRI) Images by Applying Machine Learning Techniques

Abstract

Prostate cancer (PCa) accounted for 7.8% of all new cases and was the fourth most common cancer in 2020 with 1.4 million new cases. With 15.4% of all newly diagnosed cases in 2020 being prostate cancer, it was the second most prevalent type of cancer in men globally. Due to complex nature of PCa, it is matter of concern that development of Computer Aided Diagnosis (CAD) systems for detection PCa is not keeping up with other cancer disciplines. Feature extraction using region of interest (ROI) is an important step for developing CAD systems. Around the centre of 112 PCa lesions from 99 patients, region of interest was extracted from BVAL, ADC, and T2W MRI images. Features based on two and three dimensions are extracted from the ROI. Total 444 features are extracted and used for machine learning based classification. Comparison of the proposed approach for feature extraction is tested on three classifiers viz. Support Vector Machine (SVM), Naïve Bayes (NB) and k-Nearest Neighbour (KNN). The assessment measures used to compare the aforementioned classifiers include accuracy, recall, precision, and accuracy as well as the F1-score, Receiver Operating Characteristics Curve (ROC), AUC, and U. Kappa. SVM classification outperform as best model with features extracted from ADC and T2W modality with an accuracy of 44.64 %, FPR 0.1604, and PPVGG>1 = 0.7500.