Co-registration of CTA coronary artery/vein maps and MR myocardial viability/scar maps for optimized revascularization or resynchronization therapy planning

For optimized therapeutic yield from revascularization in patients with coronary artery/ischemic heart disease, establishment of the direct anatomic relationship between coronary arterial branches and segments of left ventricular (LV) myocardium under consideration would be helpful; intervention on a branch supplying nonviable myocardium could be better avoided. Similarly, in ischemic heart failure patients being considered for cardiac resynchronization therapy, lead placement into a coronary vein tributary overlying scarred myocardium could be better avoided with prior demonstration of the relationship between coronary venous anatomy and myocardial scar distribution. Currently, the most robust modalities for noninvasive coronary artery/vein mapping and myocardial viability/scar mapping are CT angiography (CTA) and MR, respectively. We report a methodology involving the segmentation of coronary vessels and the co-registration of CTA coronary artery and/or vein maps with MR myocardial viability or scar maps.

Analysis is composed of three parts: (1) segmentation of the vessels in CTA, (2) segmentation of the myocardial scar in the DEMR, and the (3) co-registration of the two modalities.

(1) To segment the vessels we employ a novel semi-automatic technique. Firstly, high radial symmetry positions (HSPs) in the image, which include the vascular structures' centerlines and the lobe shaped structures' centerpoints, are detected by vote-based Hough-like approach. Secondly, shape tensors for each of these HSPs are calculated via Principal Component Analysis. As the tensors expose local shape properties, HSPs that do not have vessel like structures are filtered out. Then, based on these final HSPs and their shape tensors, a graph is formulated where each HSP defines a unique graph node, and node to node connection strengths are calculated based on the shape tensor similarity metric. Next, the user is asked to identify 2 seed positions, for artery and vein respectively (the only manual part of the algorithm). Finally, the graph is partitioned into artery and vein regions based on these seed points by max-flow/minimal cut algorithm.

(2) To segment the myocardial scar, prototype software, VPT (Siemens Healthcare), was employed. After automatically segmenting and subsequently editing the myocardial borders, a threshold was computed based on the mean plus 2 standard deviations of a manually selected region of remote myocardium. A 3D polygonal model was then created by interpolating the stack of 2D image regions classified as scar.

(3) A model-based approach was utilized to register the two datasets. Models were fit to contours delineating the left ventricular contours in the CTA and DEMR. The Iterative Closest Point method was applied to align them (please see Figure 1 and Figure 2 for segmentation and registration results).

Figure 1

Figure 2

Data acquisition

A patient underwent DEMR (Sonata, SIEMENS) using single shot TrueFISP inversion recovery technique approximately 20 minutes after intravenous 0.2 mmol/kg Gd-DTPA injection. The technical data were; FOV: 285–380 mm2, TE: 1.1 ms, TR: 7 ms, Flip Angle: 50 degrees, TI: 280 ms. For the same patient, CT (Dual Source Definition, SIEMENS) scan was performed with following parameters; Rotation Time: 335 ms, B25 Kernel (Dedicated Cardiac Kernel), Slice Thickness: 0.75 mm, Temporal Resolution: 82 ms, Tube Current: 261 mAs/rot, Tube Voltage: 120 kv.

Conclusion

The direct anatomic relationship between coronary arterial branches and segments of LV myocardium under evaluation for revascularization, or between coronary vein tributaries and LV myocardial regions being considered for resynchronization device placement, can be accomplished by co-registration of CTA coronary artery and/or vein maps with MR myocardial viability or scar maps.