Prostate Segmentation via Fusing the Nested-V-net3d and V-net2d
Hakan Öcal, Necaattin Barışçı · 2019 1st International Informatics and Software Engineering Conference (UBMYK) · 2019
Prostate cancer is caused by uncontrolled growth of cells in the prostate gland. Prostate cancer, unlike benign prostate gland enlargement, is not from the center of the prostate, but from the site of the tumor to the center of the capsule. Therefore, the patient experiences urinary tract complaints at the last stage. Therefore, until the last stage, the patient does not have any finding. For this reason, Magnetic Resonance Imaging (MRI) is used in regular examinations after a certain age or in various diagnostic imaging methods of patients diagnosed with this disease. Proper localization of the prostate is an important step in assisting diagnosis and treatment, such as guiding the biopsy procedure and radiation therapy. However, manual segmentation of the prostate is tedious and time-consuming. It also varies in inter-rater evaluation. The two main challenges for correct MR prostate localization are; nonhomogeneous and inconsistent appearance around the prostate border, wide prostate shape variability in different patients. In this study, Fusing the Nested 3D Dimensional Volumetric Convolutional Neural Network (Nested-Vnet3d) and 2D Volumetric Convolutional Neural Network (V-net2d) models are compared with other V-net based models. In the training conducted on the PROMISE12 dataset, 0.92 validation dice score was achieved. This study showed that the proposed model is a robust deep learning model for prostate segmentation.