Optimization And Acceleration Of Superpixel Algorithm And Its Application In Pancreas Segmentation

Superpixel segmentation is an image preprocessing stage that can effectively enhance the performance and efficiency of the segmentation algorithm.At present the best superpixel segmentation algorithm performance is Linear Spectral Clustering(LSC),which processing speed and segmentation performance is lower.Pancreas CT image is an important reference for doctors to diagnose pancreas lesions,and the accurate segmentation of pancreas organs is a prerequisite for reliable diagnosis by doctors.Because the edge of pancreas tissue is ambiguous and highly variable,the current methods of pancreas segmentation has a low segmentation accuracy,which is not enough to satisfy the needs of doctors for diagnosis.Superpixel can effectively enhance the boundary information between organs in pancreas CT images.This thesis provides a new solution for pancreas segmentation combining the features of superpixel.Therefore,the main work in this thesis mainly includes the optimization and acceleration of superpixel algorithm and its application in pancreas segmentation.The main contents are as follows:(1)We improve the processing speed of superpixel segmentation algorithm.We propose a sampling optimization method that takes different sampling ratios according to the texture complexity of image.This method can effectively improve the speed of LSC and without reduces the segmentation performance of LSC.(2)We propose a distance measurement method of superpixel segmentation algorithm.On the basis of Manhattan distance,we propose an optimized distance measurement method.This method improves the performance of non-convex pixel set image segmentation.(3)Based on the results(1)and(2),the LSC algorithm is improved.In order to adapt to the characteristics of Medical CT image,we present a medical superpixel segmentation algorithm(MLSC)and a new region merging algorithm.We propose a new interactive segmentation algorithm of pancreas that is superpixel combining region merging.The experiment show that the method has a good segmentation effect.(4)In order to improve the performance of existing automatic pancreas segmentation method and the improve interactive pancreas segmentation method of results(3),this thesis proposes an automatic pancreas segmentation method that is superpixel combining U-Net.The experiment show that the method improves the segmentation accuracy of automatic pancreas segmentation method.We can conclude from the above research that this thesis improves the processing speed and performance of LSC algorithm based on sampling acceleration and distance measurement optimization,and we propose a pancreas segmentation based on the improved algorithm combine U-Net that improve the accuracy of pancreas segmentation.