An Accurate System for Prostate Cancer Localization from Diffusion-Weighted MRI

Abstract

This paper proposes a computer-aided diagnosis (CAD) system for localizing prostate cancer from diffusion-weighted magnetic resonance imaging (DW-MRI). This system uses DW-MRI data sets that were acquired at four b-values: 100, 200, 300, and 400 smm −2. The first step in the proposed system is prostate segmentation using a level set method. The evolution of this level set is guided not only by the intensity of the prostate voxels but also the shape prior of the prostate and the voxels spatial relationships. The second step in the proposed system calculates the apparent diffusion coefficient (ADC) maps of the prostate regions as a discriminating feature between malignant and healthy cases. These ADC maps are used in the last step of the CAD system to fine-tune a pretrained convolutional neural network (CNN) to identify the ADC maps with malignant tumors. The accuracy of the proposed system was evaluated using 40% of the ADC maps while the other 60% are used to fine-tune the pretrained CNN model. The proposed CAD system resulted in an average area under the curve (AUC) of 0.95 at the four b-values.