Analysis and Evaluation of Tumor Grade of Prostate Cancer using Multiparametric MRI

Abstract

One of the most important fields of medical research is cancer diagnosis. We are witnessing major developments in cancer diagnosis particularly in improving the accuracy and time taken for detection, whether it is for lung, brain, or prostate cancer. Prostate cancer is a major health issue worldwide and second major cause of death in men after lung cancer. To identify cancer and determine its location the volumetric scanning such as ultrasound or multiparametric magnetic resonance imaging (MPMRI) are used. Prostate segmentation and tumor detection in multiparametric magnetic resonance images is a crucial and time-consuming task. The blurred prostate boundary and large diversity in prostate shape among the patients are the main challenges in automatic prostate segmentation. newlineThe objective of this thesis is to develop a computer aided detection and diagnosis (CAD) system for accurate prostate segmentation, tumor detection and predict the grades of segmented tumors using Prostate Imaging Reporting and Data System (PIRADS) guidelines. We have introduced a transparent and meticulous feature learning framework to locate the prostate automatically in MPMRI with decent segmentation accuracy. The rule based tumor identification method is used to locate tumors accurately in segmented prostate. We have detected tumors in the peripheral zone (PZ), transition zone (TZ) and central zone (CZ) of the prostate. The performance of the proposed work is enormously tested on the dataset which contains T2-weighted images (T2W), diffusion weighted images (DWI) and apparent diffusion map (ADC) images of 236 subjects. In this study a total 218 lesions were used for analysis which includes 39 non-cancerous lesions and 179 cancerous lesions. We have obtained tumor detection accuracy of 93.2% and AUC of 0.94 by using a random forest classifier. The results yielded by the proposed algorithm are validated by two experienced radiologists.