A Study Of Algorithms For Whole Heart Registration And Automatic Segmentation Of Dual Source Cardiac CT

Cardiovascular diseases have the highest Incidence rate to induce morbidity and mortality, which has been a serious threat to human health in the world. Early quantitative diagnosis of cardiovascular disease and risk assessment play key roles in extending human expectancy life. Under the control of electrocardio signal, the heart contracts and relaxes periodically in order to pump blood to the whole body to maintain the metabolism of tissues. Now, many modern medical imaging equipments can scan the heart dynamically and provide the cardiovascular images. Especially, the double source computerized tomography (CT) can provide four dimensional (4D) heart CT imaging data which is used to observe the dynamic characteristics of heart. In addition, medical image segmentation becomes a significant tool in clinical diagnosis. If each organ area and edge are extracted from the heart and the three dimensional (3D) and 4D heart model is constructed qualitatively and quantitatively, the doctors can use them for early diagnosis and prevention of cardiovascular disease, which plays an important role in clinic. However, due to the pulsation of heart, the artifact imaging and noise are produced and interfere the cardiovascular imaging. Meanwhile, the structure of heart is so complex and the volume is so small that the segmentation becomes more difficult. Because of this, automatic segmentation of the heart is a challenging issue.In this paper, under the background of the medical image registration and segmentation research , the automatic whole heart segmentation algorithm of 3D images is studied according to the characteristics of the dual source cardiac CT, and then a 4D segmentation model is constructed. Based on the segmentation model, this paper analyzes the cardiac global static parameters preliminarily.The research work of this dissertation is mainly discussed as follows:1) An automatic whole heart segmentation based on adaptive stochastic gradient descent optimization algorithm is proposed according to the features of 3D images from dual source CT. The coarse registration using affine transformation is first used to register a floating image to a reference image. Secondly, the FFD model based on B-splines is used in the accurate registration. Finally, the label image with segmentation information is mapped to the target image using the obtained optimal spatial transformation and the required segmentation is then done. During the registration, the normalized mutual information is used as an image similarity measure in the registration and B-spline interpolation is used to compute the spatial derivative of images. To prevent the local extrema, the optimization is done using adaptive stochastic gradient descent. Additionally, we employ a multi-resolution deformation strategy to improve the accuracy and robustness of segmentation.2) The dual source CT can provide a series of 3D CT images in each phase during the whole beat cycle, from which an automatic whole heart segmentation of 4D dual source cardiac CT based on the deformation field is proposed. In the cardiac multi-phase 4D dual source CT datasets, , the deformation field from the slice of the segmented 3D datasets to the corresponding slice of the 3D dataset needing to be segmented can be obtained using an automatic whole heart segmentation based on adaptive stochastic gradient descent optimization algorithm, which is used to deform the segmented results of the segmented 3D datasets to achieve the segmentation result of the 3D dataset. So the whole 4D dual source CT dataset segmentation results are obtained.3) Based on the analysis of automatic whole heart segmentation of 4D dual source cardiac CT based on the deformation field and the segmented 3D reference template datasets which is fixed and results in the segmentation distortion, an automatic whole heart segmentation of 4D dual source cardiac CT using adaptive template update strategy is proposed. The heart anatomic structure and the characteristics of small differences between adjacent layers of dual source cardiac CT are considered in this algorithm. This algorithm regards the best diastolic phase of coronary artery imaging reconstruction as the initial registration position. When the registration starts, the deformation field from the slice of the just segmented 3D datasets to the next adjacent slice of the 3D dataset needing to be segmented are obtained. The registration continues until traversing all heart phase. Finally, the whole 4D dual source CT dataset segmentation results are obtained.4) Based on the heart segmentation model of 4D dual source cardiac CT, a preliminary function analysis for cardiac static and evaluation of segmentation was carried out for the further research.Finally, specially thanks the supports of the National Natural Science Foundation of China (No.81101130) and the Fundamental Research Funds for the Central Universities under SCUT (2009ZM0235).