Research On Full Heart Segmentation In CT Images

Cardiovascular disease has the highest mortality rate globally, with mortality and morbidity increasing year by year, which has seriously been a threat to human life Therefore, quantitative diagnosis and interventional treatment for cardiovascular disease play a significant role in improving human health. The heart is the key organ of the human cardiovascular system and modern medical imaging technology can provide a wealth of information on the structure in which CT with the adventage in imaging speed and resolution is an important tool for cardiac examination. The segmentation of the main anatomical structures of the heart in CT images is very helpful in the diagnosis and treatment of cardiovascular disease. At present, domestic and international scholars have proposed many segmentation algorithms for certain heart anatomy, however they mainly focus on the atrium, ventricle segmentation. In fact, the whole heart structure is required for cardiac surgical navigation, interventional treatment,computer-aided diagnosis applications and so on. Accurate extraction of cardiac entire region and edge for establishing a full 3-D model has crucial clinical significance. However, the complexity of cardiac structures and artifacts due to the heartbeat bring cardiac location and segmentation great difficulties, so the automatic segmentation of the heart has been a challenging issue.In this paper, the medical image segmentation and image recognition as the background, focusing on highly relevant characteristic between adjacent slices and different cardiac CT image sequences, we present an automatic model-based approach enabling the segmentation of the full heart and three-dimensional visualization. The main contents of this paper include:(1) Train a classifier to achieve automatic locating of the target heart. First, based on the similarity of the individual heart, extract Haar-Like features using the existing set of samples, and generates cascade classifier by AdaBoost algorithm and cascade algorithm for detecting the input image to automatically locate the target heart. Then, according to similar characteristics of the same individual adjacent slices, dynamically adjust the current detection result to the parameters of subsequent image, forming an adaptive multi-slices detection algorithm, which effectively improves the detection efficiency and reduce the error rate.(2) Segmentation of full heart based on Active Shape Model. First, obtain a priori knowledge of the average shape model by training sample set; then, according to rectangular region detected by cardiac classifier initialize average model and adaptive shape matching. In the matching process, regarding matching result of upper-slice as the subsequent slice matching template and iterative update until the end of the traverse achieving a set of CT image sequence matching results. At last, according to the edge feature points use Bezier curve to optimal path planning and boundary selection to obtain a smooth segmentation results.(3) The results of the segmentation algorithm respectively with manual segmentation results and the clinical results, we do a qualitative and quantitative analysis, at the same time realize the three-dimensional rendering and visualization using 3D texture mapping method.In this paper, basing on a priori model automatic achieve the location of the heart shape and contour fitting match, although the pre-training model spent some time, but the results show that our study can realize an automatic and accurate segmentation of full heart, for subsequent analysis of cardiac function parameters laid the foundation data.