A fully automated system was developed to reconstruct the 3-D imagery of blood vessels from two orthogonally projected angiograms. The system is applicable to diagnostics of blood vessel abnormalities and planning of clinical procedures such as coronary angioplasty and brain surgery. Angiograms, X-ray motion pictures of blood vessels, allow the medical experts to visualize the 2-D vascular structures. The biplane angiographic system, now available in many cardiac catheterization clinics, provides simultaneous views projected from two different angles. Computerized identification of 3-D vascular structures from two views can greatly enhance the capability for clinical diagnostics of cardiovasuclar diseases; however, it is by no means a trivial task. Ambiguities of the 3-D position of vessel segments may exist and render the mathematical solution to the 2-view reconstruction problem non-unique. In this research we solved the 2-view reconstruction problem with a 2-level hierarchical approach: At the lower level, image processing techniques were used to delineate the 2-D vascular networks in each view plane. At the higher level, a rule-based system originally developed for artificial intelligence research was used to guide the matching of vessel segments between the two views. It has been demonstrated that the computerized system is capable of reconstructing 3-D vessel imagery from biplane angiograms in a fully automated way. This research has contributed to 3-D blood vessel imaging both in theory and in developing practical computer software for clinical implementation. The above figure shows the 3-D image of a left coronary arterial tree reconstructed from two orthogonally projected X-ray angiograms. The reconstruction software has been developed in the C language and under an expert system shell called C Language Integrated Production System (CLIPS). The typical reconstruction process takes less than 30 seconds on a workstation. The accuracy for the position of reconstructed vessel is within 0.8 mm.
This work was sponsored by the National Science Foundation (grant no. BCS-8910188) and conducted at the Department of Electrical & Computer Engineering of the University of Rhode Island from 1989 to 1992.