Research On Space-time Segmentation Algorithm For Pelvic Structures In CT Image Sequences

Rapid and accurate extraction of the pelvic extern contours from CT images can assists physicians in the early diagnosis of pelvic injury,in planning operations,and in evaluating the effects of surgical treatment.To solve the problem that pelvic segmentation is time-consuming and low accuracy,this paper proposed a novel segmentation algorithm based on key frame dimension reduction.The proposed method consisted of three main parts: the extraction of key frame image from the whole CT sequence,the extraction of pelvic extern contours from key frames,and the segmentation of pelvic CT image sequences.Considering that skeletal morphology was similar in two adjacent frames,a key frame extraction algorithm for CT image sequences was proposed.In this algorithm,candidate key frames were extracted from CT sequences based on pixel difference.Then,target key frames were extracted from candidate key frames based on mutual information and normalized correlation coefficient.For a single CT image,the traditional segmentation method is to manually draw the contour of the target completely by experts.Although this method can ensure accuracy to some extent,it is time-consuming and labor-consuming,which is not conducive to rapid clinical analysis.In addition,existed automatic algorithm cannot meet the clinical requirements of accuracy.Therefore,interactive segmentation method is the first choice.In order to realize the accuracy and efficiency of segmentation,the watershed algorithm based on artificial marking was used to obtain pelvic extern contours of key frames.For a CT sequence,an algorithm for extracting the pelvic extern contour was proposed which was based on the key frame segmentation information.In this algorithm,using the key frames to separate CT sequences was the first step.And determining starting contours was the second step.For CT images in each part,the starting contours of the image were determined by using the extern contours of the previous images according to the similarity between frames.Then locating the region of interest was the third step.The last step was updated the position of the contour points by analyzing the variations of each points on the starting contours in the region of interest.Through the above steps,the pelvic extern contours of all CT images can be obtained.In this paper,10 sets of CT data were used to test performance of the proposed method.And each CT sequence included 245 CT images.The results showed that approximately 13% of the images were selected as key frames,and for each sequences the average processing time was approximately 8 minutes.Take the mean of manual segmentation results by 5 experts as the ground truth to evaluate the segmentation results of the algorithm.The results showed that the average overlapping area between the results of the proposed algorithm and the ground truth was higher than 94%.Above all,the proposed algorithm in this paper can guarantee the accuracy of segmentation,shorten the time of segmentation and reduce the workload of clinical manual marking.Segmentation results not only provide an important reference for early diagnosis and decisions regarding surgical procedures,they also offer more accurate data for medical image registration,recognition and three-dimensional reconstruction.