Room: Exhibit Hall | Forum 6
Purpose: Clinical image pairs provide the most realistic test data for image registration evaluation. However, the underlying ground truth is unknown. Using combinatorial affine registration optimization (CARO) we demonstrate a method to estimate the true alignment for rigid-registration of clinical image pairs
Methods: Landmark pairs were manually selected for each CT/CBCT image pair for 6 cases representing head and neck, thoracic, and pelvic anatomic regions. From a landmark pairs, combination subsets of a k number of landmark pairs (k-combination set) were generated without repeat to form a large set of k-combination sets (k-set) for k=4,8,12. The rigid registration transformation between the image pairs was calculated for each k-combination set (24,000-300,000). The mean and standard deviation of these transformations were used to derive final registration for each k-set. Joint histogram and joint entropy analysis were used to measure and compare the quality of CARO with commercially available registration software.
Results: An average of 166 (range: 125-210) landmark pairs were selected for each CT/CBCT image pair. The mean standard deviation of registration output decreased as the k-size increased for all cases. The joint entropy evaluated for each k-set of each case was smaller than those from two commercially available registration programs indicating a stronger correlation between the image pair after CARO was used. As further proof of the efficacy of CARO the joint entropy of each member k=4 composed of 30,000 k-combination sets were calculated for one of the thoracic cases. The minimum joint entropy was found to exist at the estimated mean of registration indicating the CARO approach converges to the optimal rigid-registration results.
Conclusion: The results demonstrate the strength of CARO for estimating underlying truth for rigid-body registration between clinical image pairs. The resulting estimated truth was found to have lower joint entropy than commercial registration results.
Not Applicable / None Entered.