Research On Kidney Segmentation Of Micro-CT Images Based On Multi-atlas Registration And Random Forests

The application of medical imaging in clinical diagnosis and treatment is increasingly widespread.The segmentation of medical images provides doctors with more intuitive and effective anatomical structures,greatly improving the efficiency of diagnosis and treatment.However,medical images often have characteristics such as non-uniform grayscale,blurred boundary,and large individual differences,especially the soft tissue organs,which brings great challenges to segmentation.This article extensively investigated and analyzed the research status at home and abroad,and proposed a segmentation method based on multi-atlas registration and random forest,aiming at obtaining a clear boundary of soft tissue organ efficiently.This is a medical image segmentation framework that combines multi-atlas technology and supervised learning.Firstly,the preliminary segmentation or probability map of the target organ is obtained based on the rigid registration results of multiple atlases.Secondly,the schemes of feature extraction and classifier training making full use of the atlas information are designed for the purpose of significantly improving the segmentation efficiency,and the fine segmentation is accomplished based on the probability map by using a random forest classifier.Finally,the probability map and the classification result are integrated to generate the final segmentation.This article applied the above method to the kidney segmentation based on mirco-CT images and comprehensively analyzed the influences of different parameters on its performance.The experiments show that the proposed method achieved Dice coefficients of 0.9766 and 0.9255,mean surface distance distance of 1.25 mm and 0.98 mm on two different datasets respectively,which are obviously superior to the other algorithms.In addition,compared with related methods,this method has higher segmentation efficiency.The results proved that the proposed method integrated multi-atlas registration and random forest effectively.The coarse-to-fine segmentation strategy greatly reduces the requirements on the number of the atlases and the accuracy of the registration,and the full use of atlas information makes the segmentation more efficient.In short,this method can accurately and efficiently achieve the segmentation of the soft tissue organ in medical images,which has great clinical application potential.