The Research On Segmentation Method Of Pharyngeal Swab Image And Fetal Ultrasound Image Based On Improved U-Net

When the robot is used for pharyngeal swab sampling,the M-region in the patient’s oral cavity will play an important role in guiding the robot sampling.Therefore,it is necessary to accurately segment the M-region in the pharyngeal swab image.Similar to the research work of segmented target area in pharyngeal swab images,when the computer-aided system is used to assist sonographers in measuring fetal head circumference in ultrasonic images,the fetal head region in ultrasonic images should be segmented firstly,and the segmentation accuracy will directly affect the fitting accuracy of fetal head edge and the measurement accuracy of head circumference in the later stage.In this paper,the automatic segmentation methods of target region in pharyngeal swab image and ultrasonic fetal image were studied,and the following research works were done:Firstly,aiming at the problem that blurred or discontinuous edges of the target region affect the segmentation accuracy in the pharyngeal swab image and fetal ultrasound image,in this paper the U-Net model was improved,the expansion convolution is used to replace the ordinary convolution in the original U-Net model to obtain the multi-scale features of the image.And we design a multi-scale feature fusion module with channel attention mechanism.This module can automatically select the optimal scale features of image to improve the segmentation accuracy of the edge of the target region.This improved U-Net model was used to segment the M-region in the pharyngeal swab image.In order to alleviate the imbalance of positive and negative samples in the image,a mixed loss function was proposed to segment the M-region in the pharyngeal swab image.Secondly,in order to solve the problems in ultrasonic fetal image,such as large amount of speckle noise,interference of other tissue structure and low contrast between target area and background area,which affect the segmentation accuracy,this paper adds attention gate module on the basis of the improved U-Net model above.In order to train the model adequately,data enhancement was performed on the pharyngeal swab image dataset and ultrasonic fetal head image dataset.Finally,the segmentation performance of the improved model was verified by experiments.Different model training strategies were used for the two segmentation tasks,and Dice similarity coefficient,precision rate and recall rate were used as evaluation indexes for segmentation results.In the segmentation task of M-region,the final segmentation precision indexes of the improved U-Net model can reach: Dice similarity coefficient is 91.35%,precision rate is 93.59%,and recall rate is 88.87%.In the segmentation task of fetal head,the final segmentation precision indexes of the improved U-Net model can reach: Dice similarity coefficient is 93.27%,precision rate is 94.71%,and recall rate is 91.32%.The experimental results show that the modified U-Net model can effectively segment the target region in pharyngeal swab image and ultrasonic fetal head image.