| As a common pediatric disease, congenital heart disease is the main cause of infant death. In clinical diagnosis, most of the existing physical examinations based on the traditional two-dimensional ultrasound images, which can not provide the full detail information of the diseases.To improving the diagnostic accuracy and reliability, we design and develop a virtual endoscopy system, base on the virtual reality technique, and attempt to apply the system in the assistant diagnosis of congenital heart disease. Our research is collaborated with Shanghai Children’s Medical Center and Shanghai Xinhua Hospital, and the purpose is to investigate the implementation of applying the virtual reality technique in the qualitative and quantitative diagnosis of children’s congenital heart diseases.Speckle noise is a natural property in medical ultrasound images, which often result in blurred image features and brings unfavorable effects on the quantitative analysis and clinical diagnosis. Thus the image pre-processing operation makes significant impacts on the visual quality of the visualization results.This thesis is mainly included the following four primary parts:(1) For the image denoising problem, a new multiscale thresholding method for ultrasound image despeckling was presented by utilizing the multi-resolution properties and edge preservation feature of the curvelet transform. We also introduce and adaptive threshold estimator, by combining the stein’s unbiased risk estimate principle with the neighboring coefficients, to determine the optimal threshold. Further, the nonlinear diffusion approach was adopted to reduce the negative effects of the artifacts. The results show that our proposed algorithm can successfully to keep the balance between the speckle removal and detail edge preservation.(2) In this part, we adopt a novel image segmentation algorithm based on the combination of the conception of topological derivative and the statistical region merging strategy. In our research, a pre-segmentation process is performed on the original image by using statistical region merging method such that the whole original image domain were divided into a series of homogeneous regions. Then, the results of pre-segmentation procedure were further process by the topological asymptotic optimization process. And we apply the suggested algorithm to the implementation of medical ultrasound image segmentation. The results demonstrate our method can achieve a better segmentation result while preserving the detail information of the valves areas.(3) In order to eliminate the artifacts cause by the patients’ motion, a three dimensional rigid transformation-based automatic image registration algorithm is introduced in our research. In the experiments, we use a new similarity measure by combining the regional mutual information with the gradient information of the images. And then, the particle swarm optimization algorithm was adopted as the search strategy. The experimental results show that our registration algorithm can significantly remove the motion artifacts of the ultrasound images.(4) To provide more comprehensive diagnostic information, we expanded the basic function of the existing virtual endoscopy system by introducing a multi-view demonstration and a three dimensional measurement function. In this part, we utilized our virtual endoscopy system in the three-dimensional visualization of16patients’ cardiac ultrasound images. In addition, we measured the maximum, minimum and area of the ASD regions of all the patients. Furthermore, all the measurement values were compared with the results obtaining from the three-dimensional echocardiography. The comparison results demonstrated that there is a high degree correlation between the measurement values from the two different approaches. In conclusion, our virtual endoscopy system can improve the diagnostic accuracy and reliability, and can be applied in the qualitative and quantitative diagnosis of children’s congenital heart diseases. |
