Segmentation of Prostate Cancer in MRI Using Deep Learning

Abstract

Prostate cancer is one of the most causes of cancer deaths for men worldwide and it is still a significant public health problem. However,  prostate cancer can be cured if detected at an early stage. Due to its ability to produce detailed anatomical structures, Magnetic  Resonance Imaging (MRI) is one of the most used modalities for prostate cancer diagnosis and treatment. The accurate segmentation of  the prostate from MRI is crucial for diagnosis and treatment planning of prostate cancer. Deep learning has provided important support  in early disease detection, and image processing analysis, especially in image classification, image registration, image segmentation and medical treatment plans. In this paper, we propose an automatic segmentation of the prostate in MRI based on deep learning methods.  A convolutional Neural Network special type named 3D UNet is used to segment the prostate in MRI. We conducted the 10-fold cross- validation experiments on the public Promise 12 Data set of 50 prostate images and achieved a mean Dice similarity coefficient of 84.92%  and a mean Hausdorff distance of 5.3 mm. The experiments proved that the proposed algorithm performed with promising results.