3D Segmentation Of Cardiac Ultrasonic Image Based On Deformable Models

Medical ultrasound imaging has become an important means for diagnosis of cardiac diseases due to its characteristics of noninvasiveness, real-time and low cost, etc. To accurately obtain the integral information of cardiac cavities, the doctors need to segment the target cavities based on each image slice which has a great deal of background tissues and speckle noises, and to reconstruct the 3D heart model in their minds. That requires the doctors should have abundant clinic medical experiences and precise spatial sense. Moreover, the diagnostic results also come from the subjective judgments of the doctors. To overcome such subjectivity, we hope to represent directly by certain means the spatial positions and the shapes of cardiac cavities at different time, i.e. full 3D segmentation of cardiac cavities. Recently the techniques on the 3D segmentation of cardiac ultrasound image are still underdeveloped in the world.In this study, the more effective technique on 3D segmentation of cardiac ultrasound images is explored.The deformable models have been widely used in medical image segmentation in that they can combine effectively high-layer visual information and low-layer image information. In this study, the combination of ellipsoid initial model and 3D GVF deformable models is utilized to carry out 3D segmentation of ultrasound cardiac images. Firstly, the cardiac ultrasound volume data is matched with the ellipsoid initial model to get the ellipsoid initial contour. Secondly, the initial ellipsoid model is discretized into the triangular grid model according to certain criterion. Finally, the direct 3D deformation is adopted for the obtained ellipsoid initial contour to get the final segmentation result, through the introduced improved 3D GVF external force field. The surface rendering is conducted for the obtained segmentation result of cardiac cavities with OpenGL to achieve the 3D display of the segmentation result. The 2D projection of the 3D segmentation result of ultrasound cardiac images is carried out on parallel cross sections. The volume of the cardiac cavities is quantitatively computed through the progression of the 2D projected areas of the segmentation results.All the experimental data, the rotational transesophageal echocardiography (TEE), comes from the Cardiac Outpatients of the First Huaxi Hospital, Sichuan University. The volume data for 3D segmentation is acquired through interpolation to TEE. In the experiment, the majority of the segmentation results fit the boundary of the ventricle, and the information of cardiac cavities such as size, shape and location, etc. can be represented approximately. Compared with the segmentation result acquired through the traditional external force deformable models based on ellipse growing, the proposed model can approach the concave boundary more accurately, and the integral modality is closer to actual ventricle contours. The segmentation result by the proposed model approaches the manual segmentation result; certain difference in detail, however, exists between them.It can be concluded from the experimental results that: 1. The ellipsoid model can be used as the initial contour for 3D segmentation of cardiac ultrasound images, avoiding the subjectivity and complexity in manually selecting the initial contour; and 2. The combination of ellipsoid initial model and 3D GVF deformable models eliminates the subjectivity and complexity existing in manual selection of initial contours, and can provide satisfactory segmentation result for 3D segmentation of ultrasound cardiac images. The study provides a method for 3D segmentation of cardiac cavities from a new standpoint and presents a new approach to 3D direct segmentation of ultrasound cardiac images.