Automated prostate cancer localization with MRI without the need of manually extracted peripheral zone

In this paper, a new method that incorporates the spatial information to localize prostate cancer with magnetic resonance imaging (MRI) is proposed. Most automated methods for tumor localization require manual peripheral zone extraction from the prostate gland, and it is a tedious and time-consuming job with considerable inter-observer variability. In order to conquer this difficulty, we propose to introduce a new feature named location map to incorporate the spatial information of prostate cancer. This new feature is constructed by applying a non-linear transformation to the spatial position coordinates of each pixel, so that the location map implicitly represents the geometric position of each pixel with respect to the prostate region. Then, the location map is combined with MR images to perform segmentation. The proposed method enables us to localize prostate cancer without the need of manual extraction of the peripheral zone. Our experimental results show that the segmentation performance of the proposed method for tumors located in the peripheral zone is comparable with performance when the masks of peripheral zone are provided.