Segmentation and classification of cancer in multi parametric prostate MRI

Abstract

According to World Health Organization (WHO), Prostate cancer (PCa) has highest mortality rate in Asian continent. It is second dominant source of death among American men with an average diagnosis age of 66 years as reported by American Cancer Society (ACS) in Cancer Facts and Figures 2021. Early detection and diagnosis is the prime need of today s healthcare system, which could help in slashing the alarming mortality rate. Computer Aided Diagnosis (CAD) system has a vital role in early detection and diagnosis of cancer and acts as a reliable second observer in clinical decisions. In this work, CAD system is developed with comprehensive evaluations of different modules named as image denoising, segmentation and classification of cancer in multi-parametric (mp) prostate Magnetic Resonance Imaging (MRI). It is concluded that Anisotropic and Non-Local Means (NLM) are best denoising filter for removing noise from mp-MRI images. Proposed segmentation methodologies (I) and (II) shows comparable outcomes in terms of accuracy for delineating cancerous region in prostate gland. It is further deduced that T2w features are most contributing ones in classification of lesions as compared to other features of mp-MRI images. The presented research work is likely to contribute in clinical applications and help the urologist, oncologist and radiologist for precise, timely and prompt conclusions. newline