| In recent years,with the development of living standards,social groups pay more and more attention to their own health problems.As a convenient and non-invasive screening method,ultrasound has been broader used,the related problems of ultrasound imaging have also become a very valuable research hotspot.Due to the imaging system,manual operation and other reasons,the imaging process of ultrasound image will lead to speckle noise and uneven contrast in the image.These problems seriously limit its development in the field of artificial intelligence medical image segmentation.Different from other medical imaging scenes,ultrasound equipment is more hand-held equipment.The boundary of the image is unclear and the noise interference is large,which brings great challenges to the segmentation task.At the same time,it is not conducive to the recognition and judgment of doctors.In order to solve the above problems,reduce the workload of doctors,and meet the needs of rapid and efficient clinical diagnosis,the topic of the subject is finally determined.The research method of this topic is a universal method for the application of ultrasound image detection.We select the brachial plexus ultrasound which is difficult in clinical diagnosis to carry out the experiment,and use the brachial plexus ultrasound image(BP)data set provided in the kaggle competition.The research is divided into two parts:the first stage is to enhance the ultrasound image data set and expand the ultrasound data set.Then the following research is carried out:firstly,aiming at the shortcomings of traditional image segmentation methods,the target region is segmented by using FCN and UNet benchmark neural network,and then the edge detail information is optimized by using deep lab series models through hole convolution structure and full connection CRF to test the segmentation results.Finally,it is optimized based on deep lab V3+ model by adding self attention mechanism and modifying self attention mechanism The overall structure is optimized by changing the loss function and other strategies,and the segmentation effect of 0.7314 is finally obtained.The effectiveness of the algorithm is verified by comparing with the segmentation effect of a variety of segmentation networks.In the second stage,for the brachial plexus ultrasound image,the traditional filtering and histogram enhancement methods are used for image processing,and then the ultrasound image is processed based on the adaptive contrast enhancement algorithm.Finally,image enhancement algorithm based on image fusion is carried out on this basis,and the image enhancement effect is evaluated and compared with different standards to verify the proposed image enhancement algorithm Objective to evaluate the effectiveness and practical application value of this method in clinical diagnosis.In this paper,the effect of the segmentation algorithm of specific target position and the enhancement algorithm of ultrasonic image are evaluated respectively from the objective index and subjective visual effect,which verifies the practical value of the model in clinical application,achieves the cross application of deep learning and medical image processing,and proves its wide application prospect. |
