Multi-Channel 3D Deep Learning Architectures for Evaluation of Prostate Lesion Detection

Abstract

The localization of prostate cancer on MR images is of paramount importance for accurate diagnosis and treatment. In the present study, transformations of 3 Deep Learning segmentation models from 2D to 3D space are proposed to segment prostate lesions in MR images. The 3D Use-Net model is the best model outperforming the second by 10% Dice Score and 1.79mm Hausdorff Distance.