Detailed Investigation of Lumen-Based Tomographic Co-Registration

Imaging procedures are fundamental to disease diagnosis and prognosis, along with intervention evaluation and monitoring progression. Quantitative comparison and integration of information from different image sets require image coregistration, a process typically performed manually by clinicians. Automated and semi-automated procedures rely upon features that require extensive pre-processing or may not be visible and, as such, are not universally applicable across acquisitions and modalities, limiting applicability. We present a simple yet reliable fully-automated method that registers tomographic vascular images based only on lumen contours. The results both in multimodal and pre-/post-intervention datasets demonstrated the flexibility and robustness of our method. Excellent agreement was found between method results and manual longitudinal coregistration ($0.23\pm 0.20$ mm). Rotational co-registration accuracy was also high $(8.1^{\circ}\pm 8.0^{\circ})$, but required visible fiduciary landmarks; in the absence of such, accuracy decreased to $61.3^{\circ}\pm 66.4^{\circ}$. Sensitivity analyses indicate that, while care must be taken when overlapping segments are very short or lacking variability, lumen-based vascular image co-registration offers a robust, straightforward, and widely applicable approach.