Two-dimensional echocardiograms of the left ventricle of the heart have been used to produce three-dimensional wire frame reconstructions of the inner wall of the left ventricle chamber. Currently, this process is accomplished by tracing the boundary of the images manually after applying simple image processing algorithms. This manual interaction is a very tedious and time consuming step in the three-dimensional and four-dimensional reconstruction process. It would be desirable to remove as much operator interaction as possible from the process to improve repeatability and decrease throughput time.
This thesis investigated and implemented several basic and novel image processing algorithms which automatically preprocessed the echo images and extracted the boundary information necessary for a reconstruction process. Among the algorithms investigated are common spatial kernel operators such as the Gaussian Convolution Kernel and the Standard Deviation Kernel. Also implemented and discussed are the Robust Automatic Threshold Selector (RATS), a Contour Following Algorithm and a variation of a QUAD-TREE/Pyramid Resolution data structure. Morphological operations of erosion and dilation were applied to give the closing effect necessary in handling boundary dropouts found in some echo images. An effort has been made to utilize the Gould/Deanza image processing system to execute the algorithms in order to take advantage of its unique architecture.
The accuracy of the final sequence of algorithms was tested on actual images and a comparison made between hand drawn borders and the computer generated borders traced on actual echo images.
Computer Science (MS)
Department, Program, or Center
Computer Science (GCCIS)
Raqueño, Rolando, "Automated boundary detection of echocardiograms" (1990). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus