The objective of this research is to extend the clinical utility of three-dimensional echocardiography (3DE) by developing automated means for left ventricular (LV) function estimation. This dissertation presents our work on a semi-automated algorithm that, without extensive supervision, tracks the LV boundary through the spatial and temporal sequences of two-dimensional frames that constitute a 3DE data set. The construction of the algorithm is based on a framework of factor graph representations and probability propagation. One key component is the derivation of "LV edge likelihoods" as image features that provide "soft" edge information rather than a "hard" edge map to make LV tracking more robust to frame-to-frame variations in feature sizes and intensity levels. The algorithm begins with the operator marking some highly visible landmark points along the LV boundary in a few spatially separated frames. This takes a few seconds to complete, and is the only operator input to initiate the procedure. Full boundary estimates in these initial frames are completed by spline fitting to the selected points. Using spatial continuity in the LV boundary, these estimates establish search regions for the intermediate frames, within which boundary points are specified as those having highest edge likelihood. A similar procedure, employing forward and backward tracking, is used for the temporal sequence of frames at each spatial location. LV volume as a function of time is calculated from the set of estimated boundaries using a modified version of planimetry. Our system's performance is tested on gated-rotational and real-time 3DE data from both normal and diseased hearts, obtained using Philips ultrasound systems. The results are validated by comparison to the "gold standard" of nuclear scan (Single Photon Emission Computed Tomography or SPECT) volumes for the same hearts. As a critical part of our algorithm, the development of a method for characterizing the mean square error (MSE) in LV volume estimates is presented, which serves as the quality indicator that can flag unreliable estimates. The utility of this self-verification information is also validated as part of the nuclear scan comparison studies. Preliminary results show that it does seem to distinguish between more and less reliable estimates.
Identifer | oai:union.ndltd.org:UMASS/oai:scholarworks.umass.edu:dissertations-4628 |
Date | 01 January 2007 |
Creators | Shin, Il-Seop |
Publisher | ScholarWorks@UMass Amherst |
Source Sets | University of Massachusetts, Amherst |
Language | English |
Detected Language | English |
Type | text |
Source | Doctoral Dissertations Available from Proquest |
Page generated in 0.0161 seconds