Deep Guidance Network for Biomedical Image Segmentation

Pengshuai Yin,Qingyao Wu,Rui Yuan,Yiming Cheng

doi:10.1109/access.2020.3002835

Abstract

Segmentation of 2D images is a fundamental problem for biomedical image analysis. The most widely used architecture for biomedical image segmentation is U-Net. U-Net introduces skip-connections to restore the spatial information loss caused by down-sampling operations. However, for some tasks such as the retinal vessel segmentation, the loss information of structure can not be fully recovered since the vessels is merely a curve line that can not be detected after several convolutions. In this paper, we introduce a deep guidance network to segment the biomedical image. Our proposed network consists of a guided image filter module to restore the structure information through the guidance image. Our method enables end to end training and fast inference (43ms for one image). We conduct extensive experiments for the task of vessel segmentation and optic disc and cup segmentation. The experiments on four publicly available datasets: ORIGA, REFUGE, DRIVE, and CHASEDB1 verify the effectiveness of our method.

Highlights

Deep neural networks especially convolutional neural networks (CNNs) outperform the state-of-the-art in many visual recognition tasks
EXPERIMENTS We evaluate the model for two tasks: Optic disc and cup segmentation and retinal vessel segmentation
False negatives (FN) are the misclassifications where a vessel pixel in the ground truth image is classified as non-vessel in the segmented image and the false positives (FP) are the misclassifications where a non-vessel pixel in the ground truth image is marked as the vessel in the segmented image

Summary

INTRODUCTION

Deep neural networks especially convolutional neural networks (CNNs) outperform the state-of-the-art in many visual recognition tasks. 2. The backbone of our network is U-Net. We design a guided image filter module to preserve edge information and reduce the noise effect of the feature map. 2) MULTISCALE GUIDED FILTER MODULE The down-sample operation of the U-Net results in the spatial information loss especially for the tiny thin vessels in the retinal image, which cannot be restored through skip-connections or up-sample operation. For optic disc and cup segmentation, the boundary between the optic cup and optic disc is weak To solve this problem, we design a guided filter module to preserve the edge information from the grey-scale guidance image. Guided filtering module takes feature map F and guidance grey-scale image G as inputs, generating the output O. The final classifier treats the segmentation as the pixel-wise classification to produce the probability map at each pixel

GUIDED FILTER MODULE

EXPERIMENTS

RETINAL VESSEL SEGMENTATION

CONCLUSION