Research And Realization Of Interactive Segentation Methods On Medical CT-Image Sequences

Medical images are so particular and complex that the general image segmentation methods cannot obtain ideal results. Interactive image segmentation methods can integrate subjective experiences into the image segmentation process so that they have irreplaceable advantages over other image segmentation methods. As medical images often have an amount of CT-images, complete manual segmentation method cannot meet the clinical needs. Thus, the study focus on both minimizing the workload of physicians and integrating the subjective experience into the process of medical image segmentation.First, the thought of interactive contour approximation is supposed as a foundation. Then, combined with traditional segmentation methods, such as the regional similarities and the edge detection algorithm, some new interactive segmentation algorithms are studied. The content mainly includes:(1) Introduce the concept of the sub-regions similarity into the method of geometrical contour approximating, a new polygon contour approximating algorithms based on the sub-regions similarity is proposed. The experiments show that the algorithm based on sub-regions similarity can improve noise immunity in the process of the contours evolution. Meanwhile, the different styles of the initial contours can meet to different demands from practice.(2) Combined the edge detection technology using the maximum wavelet modulus and the contour approximating methods, the new image segmentation algorithm is presented. The experiments show that the edge detection algorithm's capability of noise rejection can be improved and the result of edge detection can provide great base for the process of contour approximating later. The method can achieve good results for different practical demands and offer the flexibility to the interactive process.(3) Taking Microsoft VC++6.0 as development environment, an interactive segmentation system for medical CT Image sequences has been realized according to the above two kinds of interactive segmentation algorithms. First, the initial polygon contours are obtained through physicians'interactive acts for several representative CT-images. Then, with the projection technology, the contours of whole CT-images are also attained. Finally, the target areas are distinguished from the CT-image sequence taking advantage of the contour approximating algorithm. The system is flexible and practical.