X-ray angiography (XA) is the gold standard for capturing real-time vessel structure dynamics as well as catheter location during percutaneous coronary intervention (PCI). XA can further be utilized to accurately quantify the severity of coronary sten...
X-ray angiography (XA) is the gold standard for capturing real-time vessel structure dynamics as well as catheter location during percutaneous coronary intervention (PCI). XA can further be utilized to accurately quantify the severity of coronary stenosis. However, partial vessel structures are visible for only 20 s after a contrast medium is injected, and the projective nature of angiography makes it difficult for a physician to quickly and accurately recognize a patient’s cardiovascular shape.
In order to overcome these challenges, a pre-operative 3D computed tomography angiography (CTA) image can be combined with an intra-operative 2D XA image to provide a road map of previously invisible blood vessels by visualizing the selected vascular structure in the CTA image. This technique of reconstructing complementary information obtained from different images into one space is termed “image registration.” It enables information about the anatomical structure gained from the CTA image, and required for XA management, to be transferred into XA for a patient's emergency care record.
In this paper, a feature-based 3D/2D non-rigid registration using information regarding coronary anatomy is proposed for aligning an intra-operative XA image with pre-operative CTA image. The proposed methodology is composed of two parts: in the first, initial registration calculates the global transformation of each sub-structure and selects a suitable sub-structure of the entire 3D vascular structure for a given XA image; in the second, a spline-based non-rigid registration method is used to compensate for local shape discrepancies due to heartbeat and respiration. In both cases, the registration algorithms are based on vessel centerlines. To automatically extract vessel centerlines from an XA image, a method of vessel segmentation integrated with histogram modification is also proposed.
To briefly describe the principle of the proposed approach, preprocessing is conducted by constructing local histogram modifications that reflect the intensity of neighboring pixels. This method can be used to correct non-uniform intensity along the vessels and improve the results of the vessel enhancement filter in vessel segmentation.
Next, the 3D vascular structures projected from the vessel centerlines, which were found using CTA, are divided into two sub-structures and the one best matched to the XA image is selected by using a global rigid transformation between each sub-structure and vessel centerline in XA image. To efficiently search the parameters of this global transformation, a modified version of 2D distance transformation is computed using a distance map. The selected sub-structure, using the current estimate of the optimum transformation, is the input for the non-rigid registration.
Subsequently, a thin plate spline robust point matching (TPS-RPM)–based algorithm is proposed for use as the non-rigid registration method, defining the local shape deformations that follow the initial registration. The TPS-RPM algorithm adopts the softassign algorithm and deterministic annealing technique to relax binary correspondence into fuzzy correspondence and avoid local maxima.
Finally, automatic segmentation of two ostia and a coronary anatomy labeling method are proposed. The ascending aorta is initially detected using a Hough circle transformation and geodesic active contours. The two ostia are detected within the region of the refined ascending aorta by a vessel enhancement filter. The angle in polar coordinates is then used to label the coronary anatomy.
The proposed feature-based non-rigid registration using information on the coronary anatomy can reduce projection errors when performed accurately, selecting the optimal sub-structure to be subjected non-rigid registration and providing an optimized initial position prior to this registration. In addition, spline based non-rigid registration can reflect local variations of coronary vessels. Experimental results based on 12 clinical datasets have demonstrated that the proposed method is effective, with an average distance error of 0.68±0.2 mm and an average visual evaluation score of 4.3 on a 5 point scale when rated by clinical experts. This method’s potential applications in a new field beyond the scope of diagnosis and treatment were confirmed by using the proposed method as a way to transfer coronary anatomy labeling from a CTA image to an XA image to manage patient’s emergency care record.