Research On Prostate Ultrasound Image Segmentation Algorithm Based On Semantic Constraint Expression And Bidirectional Time Series Denoising

With the rapid development of society,the busy and fast pace of work and life makes more people have hidden diseases.Among them,the disease caused by prostate is one of the diseases with the highest incidence rate in men.Among the methods for treating related diseases,transrectal ultrasound(TRUS)has emerged as a commonly used treatment method due to its speed and safety.In medical research and practice,in order to obtain pathological information about the prostate region,it is necessary to identify and measure the shape and boundary of the tissue region.However,due to the lack of prostate data and the large number of complex spots,noise,and artifacts in ultrasound images,the segmentation of TRUS images has become a challenging task in computer segmentation.The research difficulties of this topic are as follows:(1)Organ boundary blurring: Due to the impact of imaging principles and imaging equipment,TRUS image quality is not high,causing interference to computer segmentation;(2)Limitations of labeling data: Due to the requirement of experienced physicians to annotate TRUS images for prostate organs with blurred boundaries,the cost of labeling this image is high,resulting in a scarcity of labeled TRUS images;Secondly,the label is not credible.Due to the degree of patient’s cooperation during image shooting and the impact of equipment,the image information does not match the labeled data,resulting in inaccurate confidence in the general model;(3)Based on the existing initial segmentation,fast,accurate and low-cost segmentation of outliers in non significant regions is achieved.In this context,in order to achieve rapid and accurate segmentation of prostate ultrasound images.This paper proposes a bidirectional semantic constrained segmentation model for TRUS images.The main contributions of this model are as follows:(1)In order to limit the segmentation area to a certain range,a semantic constraint is proposed to represent ultrasound images,thereby unifying the pixel and shape features of the images;(2)In order to avoid the same large-scale outlier as those in the deep learning segmentation results,neighborhood information is added to the normal vector boundary operator,and the impact of neighborhood information on the segmentation results is explored;(3)An exponential time series denoising algorithm is proposed to solve outlier in a small area;(4)The combination of operator and denoising algorithm and bidirectional iteration can better solve the outlier in non significant areas and keep the effective area close to the real boundary point.Experimental results show that this model has better segmentation performance compared to classic and popular deep learning methods(U-net,Seg Net,Deep Lab V3+,Bise Net V2),with an average Dice similarity coefficient(DSC)of 96.74% and m Io U of 93.71%,achieving accurate and fast segmentation of prostate ultrasound images.