No abstract

Prostate cancer (PCa) is responsible for the maximum deaths of men across the world after lung cancer; hence, it should be diagnosed in the initial stages. Magnetic resonance imaging (MRI) commonly diagnoses PCa due to better visibility of desired organ and cancerous region over other modalities. Therefore, development of MRI-based computer-aided diagnosis (CAD) systems for PCa has become a recent area of research. Conventional methodologies used by researchers and radiologists were time consuming and prone to subjective errors due to manual interpretation. Thus, the CAD system helps in the early detection of PCa by reducing the computational complexity and increasing the detection accuracy with less chances of subjective errors. This article proposes a deep learning-based methodology named prostate cancer segmentation network (PC-SNet) for the segmentation of the region of interest (ROI) from the MRI sub-modalities T2-weighted (T2W) and dynamic contrast enhanced (DCE). Further, the performance is analyzed using parameters such as accuracy, Mathews correlation coefficient (MCC), dice similarity coefficient (DSC), Jaccard Index (JI) or intersection over union (IOU), F-score, and Hausdorff distance (HD). Finally, the performance of PC-SNet is found to outperform fully convolutional network (FCN), semantic pixel wise segmentation (SegNet), residual network (ResNet), UNet, and ENet architectures.