Research On Automatic Recognition And Segmentation Of HIFU Treated Area

High Intensity Focused Ultrasound has gradually become popular in the clinical treatment of tumors due to its advantages of minimally invasive and high efficiency.The main principle is that the ultrasonic energy is focused on the lesion area to generate a high temperature above 65℃,which causes rapid degeneration,coagulation and necrosis of the tumor tissue,but causes little damage to the tissue outside the lesion area.Therefore,in the process of HIFU treatment,the monitoring of the damaged status,shape and size of the treated area plays an important role,it can guide doctors to give a further treatment effectively.In this paper,fresh isolated pork tissues were taken as experimental samples to obtain B-ultrasound images before and after HIFU irradiationand,64×64 HIFU treated area images were intercepted to study the denaturation recognition method of HIFU treated area,and study further the segmentation method of HIFU treated area.The main work is as follows:(1)Optimization of automatic recognition method for HIFU treated area: after objective screening,the optimal combination of the selected characteristic parameters was used to identify the variability of HIFU treated area automatically by the Generalized Regression Neural Network suitable for small samples and the Probabilistic Neural Network suitable for two classification.Firstly,14 characteristic parameters from gray-gradient matrix and 4 characteristic parameters from gray difference of ultrasonic subtraction image were extracted,then using student’s t test method and Wilcoxon rank sum test to screen redundant characteristic parameters.Secondly,using the Euclidean distance determination method to select the remaining characteristic parameters again.Finally,the best characteristic parameters of inhomogeneity gray distribution and inhomogeneity gradient distribution in the front of Euclidean distance were selected and combined randomly into GRNN and PNN for automatic identification of HIFU treated area.The results show that the total recognition rate of the best combination of GRNN and PNN was 98.04%and 99.02% respectively,which was 18.63% and 33.33% higher than the average value and contrast combination of input screening.The recognition rate of HIFU treated area was improved successfully,and the best combination of GRNN and PNN was 5.88% and 6.86% higher than the SVM used in the group.(2)The shape,boundary and size of HIFU treated area were studied further.The absolute difference’s range of gray mean value between target region and background region obtained by local statistics was proposed as the rough estimation method of threshold range of region growing method,and proposing quick shift clustering algorithm combined with region growing method as a new way to segment HIFU treated area of B-ultrasound image.Firstly,quick shift algorithm was used for image clustering,then combined with region growing algorithm to segment the specific shape of treated area.The results show that: this method optimized the region growing algorithm,so that the threshold value in a certain range of any value could maintain the stability of segmentation performance,and the segmentation accuracy was also improved.