Accurate, fully-automated registration of coronary arteries for volumetric CT digital subtraction angiography

Diagnosis of coronary artery disease with Coronary Computed Tomography Angiography (CCTA) is complicated by the presence of signi cant calci cation or stents. Volumetric CT Digital Subtraction Angiography (CTDSA) has recently been shown to be e ective at overcoming these limitations. Precise registration of structures is essential as any misalignment can produce artifacts potentially inhibiting clinical interpretation of the data. The fully-automated registration method described in this paper addresses the problem by combining a dense deformation eld with rigid-body transformations where calci cations/stents are present. The method contains non-rigid and rigid components. Non-rigid registration recovers the majority of motion artifacts and produces a dense deformation eld valid over the entire scan domain. Discrete domains are identi ed in which rigid registrations very accurately align each calci cation/stent. These rigid-body transformations are combined within the immediate area of the deformation eld using a distance transform to minimize distortion of the surrounding tissue. A recent interim analysis of a clinical feasibility study evaluated reader con dence and diagnostic accuracy in conventional CCTA and CTDSA registered using this method. Conventional invasive coronary angiography was used as the reference. The study included 27 patients scanned with a second-generation 320-row CT detector in which 41 lesions were identi ed. Compared to conventional CCTA, CTDSA improved reader con dence in 13/36 (36%) of segments with severe calci cation and 3/5 (60%) of segments with coronary stents. Also, the false positive rate of CTDSA was reduced compared to conventional CCTA from 18% (24/130) to 14% (19/130).