Research Of Algorithm For Automatic Segmentation Of Tissue Region Based On Medical Sequence Images

With the development of medical imaging technology and computer technology,computer-aided medical diagnosis has been widely used.Specifically,medical image segmentation technology plays a crucial role as a key step.The result of image segmentation directly affects the correct rate of subsequent analysis.Due to the complexity and difference of human tissue structure and the high requirements of medical diagnosis for segmentation accuracy,the commonly used medical segmentation algorithms are mainly semi-automatic segmentation algorithms based on artificial participation.However,with the continuous increase of medical data,this will increase the workload of medical staff,so affecting the efficiency of diagnosis.Therefore,studying an automated segmentation algorithm has important theoretical and practical significance.In response to this problem,this paper proposes two kinds of segmentation algorithms respectively.The specific research work and innovations are as follows:1.An algorithm for automatic segmentation and reconstruction of medical images based on three-dimensional space is proposed.It is difficult to set the seed point in the traditional region growing algorithm,and the algorithm uses only two-dimensional information of image data.The problem of position information in three-dimensional space in sequence images is improved.The automatic extraction algorithm of seed points,the growth strategy combined with edge constraints,and the termination strategy based on variance between classes are proposed.The proposed algorithm is compared with the expectation maximization method of human brain tissue based on the magnetic resonance image,and the segmentation accuracy is improved by about 2%.2.A medical image segmentation network based on seed point is proposed.The algorithm is improved based on CAN(Context Aggregation Module),and the network input and network loss functions are improved respectively.The seed point image and the seed point distance transform image of the segmentation area are added in the network input stage to enhance the network input,and the seed point image is automatically extracted according to the seed point automatic extraction algorithm proposed in the paper;in the loss function stage,the network loss function is optimized,and the network is allowed.There is a possible output,and based on the minimum Tversky loss,a decrementing penalty factor is added for each result’s loss,ensuring that the first result of the network output is the least lossy result from the labeling result.The experimental results of this algorithm are compared with the image block-based convolution network and the full convolution network,and the accuracy is improvedThe algorithm of this paper is mainly used to realize the problem of automatic segmentation of medical images.Compared with the existing algorithms,the segmentation accuracy is improved,and the validity and feasibility of the proposed algorithm are verified.