Session PB7.5

Epicardial Coronary Angiography from Microbubble-Based Tridimensional Echocardiography: A Feasibility Study

DM Lage*, JM Tsutsui, SS Furuie

University of São Paulo
São Paulo, Brazil

Conventional coronary angiography is the current gold standard to evaluate coronary stenosis severity. However, this is an invasive procedure which is based on ionizing radiation (X-Ray) and dependent on nephrotoxic contrast agents. Thus, several alternative angiography techniques are being developed in the last years. Although available techniques already exist, such as magnetic resonance and computerized tomography, they require high-costs technology.
In the past three decades, echocardiography has emerged as an important medical image modality in Cardiology. Due to its easiness of use and its multiple attractive features, this technology raised significant interest in the research community. Echocardiography was endorsed by many works, most of which indicating its unique ability in different scenarios. With the advent of microbubble-based contrast agents and array transducers, 3D-echocardiography now presents itself as a relative low-cost, non-invasive and non-ionizing alternative method to visualize arteries and their dynamics.
Since available computational tools do not provide satisfactory processing, this paper investigates segmentation techniques to emphasize and isolate epicardial coronaries in tridimensional microbubble-contrasted echocardiographic images. A preliminary step was to test 4 different image segmentation algorithms based on fuzzy-connectedness theory that seems to be a suited approach for the problem. Three of them are from literature and one original contribution. Basically, the proposed approach consists in delimiting the object using a seed and a guide-seed voxels, both defined by the user. These selections provide enough data to improve the fuzzy-connectedness attributes and parameters computation, since they optimize weights estimation, fuzzy-affinity calculation and shortest-path algorithm.
We have used a training set of 240 images simulating patient condition, artery geometry, equipment setup and operator ability, 85, 5% to 92, 0% of accuracy was reached by literature methods, and 95, 2% of accuracy by our proposed method. These results reveal a significant improvement in the method and an incentive for future work in the area. The next step is to assess these algorithms using real examination images. Preliminary tests on 2 real images showed the potential usefulness of the proposed methodology.

(Abstract Control Number: 195)