Three-dimensional reconstruction of coronary arteries from multiple projections

Research is conducted to develop and evaluate quantitative techniques to assess the morphology of major coronary arteries by means of three-dimensional (3-D) reconstructions from X-ray angiography. The fundamental research hypothesis is: positions and dimensions of major coronary arteries can be obtained, to the accuracy and precision on the order of the pixel dimension, by systematically integrating geometric and densitometric information extracted from a few views of X-ray angiograms through 3-D reconstruction. With sufficient reconstruction accuracy and precision, this technique can provide global quantitative information about the coronary arteries that may not be available using other techniques.; A multi-view reconstruction algorithm is developed and is implemented on a graphics-oriented workstation. The output of the system is a 3-D tree structure, which includes vessel positions and dimensions for each branch, of the coronary arterial bed. The final reconstructed tree is obtained by iteratively incorporating into the structure 2-D geometric and densitometric vascular information form multiple views. Edge detection using matched-filtering and dynamic programming provides vessel geometric edge and centerline positions. The 2-D centerlines are used, along with the view geometry, to determine corresponding 3-D centerlines. Points in all the views are matched using the epipolar constraints and the criteria that minimize projection errors in all the views. Densitometric measurements of vessels are performed in each view. By averaging these measurements across the image, a density-length conversion factor k is determined. Absolute vessel lumen areas are computed by weighted combinations of inclination angle corrected densitometric areas which are independently obtained from all views.; The study of reconstruction accuracy and precision is difficult because the true positions and dimensions are normally unknown. A series of experiments are designed and performed including computer simulations, aluminum, wire phantom studies, pig coronary artery cast studies, and a clinical study. The phantoms used in these studies are constructed so that at least one parameter under investigation is precisely known or is carefully measured. For the clinical study, the reconstruction is compared with a clinically used manual method. It is demonstrated that subpixel accuracy can be obtained in reconstructing vessel lumen areas and pixel accuracy can be obtained in reconstructing vessel centerline positions using two or three views.