2019/02/11 by Balamurali Murugesan, Murugesan, Balamurali, Kaushik Sarveswaran +7 · 3 citations
Computer Science · Engineering · #AI in cancer detection #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Medical Image Segmentation Techniques #Medical Imaging and Analysis
paper · pdf · doi:10.48550/arxiv.1902.04099
openalex publication_date 2019/02/11 · openalex created_date 2022/07/29 · openalex updated_date 2026/07/28
Image segmentation is a primary task in many medical applications. Recently,\nmany deep networks derived from U-Net have been extensively used in various\nmedical image segmentation tasks. However, in most of the cases, networks\nsimilar to U-net produce coarse and non-smooth segmentations with lots of\ndiscontinuities. To improve and refine the performance of U-Net like networks,\nwe propose the use of parallel decoders which along with performing the mask\npredictions also perform contour prediction and distance map estimation. The\ncontour and distance map aid in ensuring smoothness in the segmentation\npredictions. To facilitate joint training of three tasks, we propose a novel\narchitecture called Psi-Net with a single encoder and three parallel decoders\n(thus having a shape of \Ψ), one decoder to learns the segmentation mask\nprediction and other two decoders to learn the auxiliary tasks of contour\ndetection and distance map estimation. The learning of these auxiliary tasks\nhelps in capturing the shape and the boundary information. We also propose a\nnew joint loss function for the proposed architecture. The loss function\nconsists of a weighted combination of Negative Log likelihood and Mean Square\nError loss. We have used two publicly available datasets: 1) Origa dataset for\nthe task of optic cup and disc segmentation and 2) Endovis segment dataset for\nthe task of polyp segmentation to evaluate our model. We have conducted\nextensive experiments using our network to show our model gives better results\nin terms of segmentation, boundary and shape metrics.\n