Efficient algorithm for the determination of left ventricular borders using cine magnetic resonance images

An efficient computerized solution is obtained for the quantitative evaluation of ventricular function using cine magnetic resonance (MR) cardiac images. This work originated from the need for a practical and reliable method of spatiotemporal surface detection. The purpose of this thesis is to develop and evaluate a flexible algorithm for extracting a sequence of contours in a spatiotemporal surface with reduced amount of human involvement. Because most existing surface detection techniques are impractical for the analysis of four-dimensional (three-dimensional (x,y,z) in time) image data, the significance of this approach is best appreciated by the physical attempt to obtain the same solution manually. Furthermore, the development is open to evolve into a better solution in the sense that the exact solution is achievable by repetitive use of the algorithm. The resultant endocardial and epicardial detected cardiac borders that are used for cardiac measurements are found to be accurate and reproducible. Because of these reproducible characteristics, the algorithm can be a useful tool to evaluate cine MR image quality. The semiautomated algorithm consists of three components: a priori target contour entry, automated contour detection, and a posteriori contour correction. The initial step captures all information about the given cardiac structure and extends throughout the image sequence. The second step is designed to exploit the temporal behavior of the tomographic cardiac structure based on time-motion analysis. The third step effectively combines the information from the previous steps with user corrections. Preliminary results using the semiautomated algorithm are found to reproduce hand-traced cardiac measurement with the accuracy of expert performance. Pixel-by-pixel comparison with the hand-traced contours has shown that the algorithm can efficiently and reproducibly generate left ventricular borders using a large number of spatiotemporal cardiac images. This experiment concludes that spatiotemporal surface detection via time-motion analysis is a simple and effective technique for the analysis of four-dimensional cardiac image data.