Medical Image Segmentation Based On UNET

With the rapid development and wide popularization of medical imaging technologies,a large number of medical images are collected and could be used for analysis.Image segmentation is one of the main research areas in medical image analysis,which attempt to assign the labels to each pixels and address the pixel-wise classification.It is emergent to develop automatic algorithms to efficiently and objectively analyze these medical images,with the aim of providing doctors with precise interpretation of diagnosis information contained in the images to have better treatment of a large amounts of patients.Medical image segmentation is a critical and important step for developing computer-aided system in clinical situations.It remains a complicated and challenging task due to the large variety of imaging modalities and different cases.In face of the above challenges,this paper address three key question:(1)The discrimination between target and background is low;(2)The problem of adhesion and aggregation between targets;(3)The problem of model lightweight design.The method proposed in this paper can achieve more detailed analysis and understanding of medical images,effectively improve the performance of medical image segmentation to a certain extent,and assist doctors to get more accurate diagnosis results.Therefore,the method proposed in this paper has important significance for accelerating the process of disease diagnosis and scientific discovery.The contributions of this thesis are mainly as follows:1.Contour-aware semantic segmentation network with spatial attention mechanism for medical imageTo solve the problem of low discrimination between target and background in medical images,a contour aware semantic segmentation network based on u-net is proposed by using the boundary and other detail information.The proposed method include a semantic branch and a detail branch.The semantic branch focuses on extracting the semantic features from shallow and deep layers,the detail branch is used to enhance the contour information implied in the shallow layers.In order to improve the representation capability of the network,a Multi Block module is designed to extract semantic information with different receptive fields.Spatial attention module(CAM)is used to adaptively suppress the redundant features.Comparison with the state-of-the-art methods,our method achieves a remarkable performance on several public medical image segmentation challenges.2.An Integration Convolutional Neural Network for Nuclei Instance SegmentationIn order to solve the problem of mutual coverage and adhesion between objects in pathological tissue images,UNET based contour aware medical image semantic segmentation network can not extract the boundary information of objects,This thesis proposes a nuclei instance segmentation method with the aim of jointing detection and segmentation simultaneously.The method builds on a two stream Convolutional Neural Network(CNN)architecture that explicit utilize a single shot multi-box detector with Feature Pyramid Network in one stream for obtaining the nuclear location information and keep an U-net in the other stream for nuclei instance segmentation.Furthermore,the recurrent,residual and attention mechanisms are integrated for focusing on the useful information.The approach utilizes the strengths of the Recurrent Convolutional Neural Network(RCNN),the Attention Network and the Residual Network.The experiments demonstrate that the proposed nuclei instance segmentation approach outperforms prior state-of-the-art methods,and could be generalized across variety of nuclear type,magnifification and imaging modality.3.A Fast and Accurate Algorithm for Nuclei Instance Segmentation in Microscopy ImagesAlthough the UNET based accurate nuclear instance segmentation network has some advantages in accuracy,its speed is greatly limited.This thesis propose a fast and accurate box-based nuclei instance segmentation method.Mainly,we employ a fusion module based on the feature pyramid network(FPN)to combine the complementary information of the shallow layers with deep layers for detection the nuclear location by bounding boxes.Subsequently,we crop the feature maps according to the bounding boxes and feed the cropped patches into an U-net architecture as a guide to separate clustered nuclei.The experiments show that the proposed approach outperforms prior state-of-the-art methods,not only on accuracy but also on speed.