Research On Medical Image Segmentation Algorithm Based On Deep Learning

Medical image segmentation is a very crucial step in medical image analysis,which aims to assist doctors in obtaining important information about diagnostic organs or tissues.It is very important for quantitative pathological assessment,treatment planning and disease progression monitoring.But medical image segmentation is still a complex and challenging task because of the various types of imaging modalities and the varying conditions of patients.With the development of deep learning techniques,medical image segmentation can be performed without relying on manually designed features,and neural networks can automatically learn the features needed for the segmentation task.As a result,deep learning-based methods have become the primary choice for researchers performing medical image segmentation.However,the existing deep learning-based medical image segmentation algorithms still have significant shortcomings,which cannot meet the current needs of medical image segmentation.To address the shortcomings of these algorithms,we propose corresponding improvement methods and verify the effectiveness of the proposed methods through experiments.The method proposed in this paper enables a more detailed analysis and understanding of medical images.It can effectively improve the performance of medical image segmentation to a certain extent and assist doctors in arriving at more accurate diagnostic results.Therefore,the method proposed in this paper is of great significance in speeding up the process of diagnosis and scientific discovery of diseases.The major points of research in this paper are as follows:(1)According to the characteristics of different medical images,we propose a pre-processing algorithm for enhancing medical images.For medical images with complex shape and structure such as retina and coronary arteries,we first extract the green channel of the image.This is followed by a normalisation process.Finally,contrast-constrained adaptive histogram equalisation and gamma transform are performed,which effectively improves the segmentation accuracy of the network.For medical images of lung,liver and other block structures,we perform the normalisation operation,which effectively improves the segmentation efficiency of the network.(2)In medical image segmentation tasks with complex shape and structure such as retina and coronary artery,the existing segmentation algorithms have problems such as insufficient feature extraction capability and low segmentation efficiency.In order to address these problems,we propose a medical image segmentation algorithm based on multi-scale feature extraction.Firstly,the algorithm replaces the original convolutional blocks in UNet with the proposed deconvolutional segmentation module,so that the network retains more detailed information.Then,the mixed pooling module(MPM)and template convolution module(TConv)are introduced in the intermediate connection section of encoding and decoding,which can improve the network’s ability to extract multi-scale features and thus improve the segmentation quality and segmentation efficiency.We experimentally validate the proposed algorithm on the DRIVE and STARE fundus datasets.The results demonstrate that the algorithm performs well in segmentation.It can effectively enhance the representation of linear features and shape features of the target,thereby making the target and background more distinguishable.(3)In the task of segmenting medical images of lung,liver and other block structures,existing segmentation algorithms have problems of segmentation breakage as well as adhesion.In order to address these problems,we propose a medical image segmentation algorithm based on an involution UNet network.Firstly,the algorithm uses a hybrid convolution-involution as the main structure of the network.Secondly,we introduce the involution operation and Inception module in the downsampling module of the network,which further extends the network width and alleviates the feature loss problem.In addition,the network can effectively improve the segmentation efficiency by introducing the Mish activation function.We experimentally validate the proposed algorithm on a lung CT dataset.The results demonstrate that the algorithm performs well in segmentation and ensures the integrity and continuity of the segmentation results.