(Tuesday, 7/16/2019) 3:45 PM - 4:15 PM
Room: Exhibit Hall | Forum 6
Purpose: The deformation vector field (DVF) is considered the ground truth in evaluating deformable image registration (DIR) accuracy. This study investigated the DVF errors and contour-based metrics for a unimodal intensity-based DIR algorithm for multiple anatomic sites on CTs with various deformation intensity levels.
Methods: Ground-truth DVFs were generated by contour-controlled deformations on three anonymized patient’s CT datasets (head-and-neck (HN), thoracic, and pelvis) using ImSimQA software (Oncology Systems Limited, UK), with three levels of deformation intensity (low, median and high) for each site. A total of nine DVFs and their corresponding image pairs (with associated contours) were imported to MiM-Maestro system. Image deformation and contour propagation were performed, followed by exporting and calculating the DVF errors and contour-based metrics, including mean distance to agreement (DTA) and dice similarity coefficient (DSC). The correlation between DVF errors (dmin, d95%, d50%, d5%, dmean and dmax) and metrics were calculated using the Pearson correlation coefficient, where R > 0.8 was considered statistically significant.
Results: Deformation accuracy analysis for HN, lung, and pelvis showed average dmean in DVF errors (21.43, 1.16 and 1.93mm), Mean DSC (0.86, 0.92 and 0.93), and mean DTA (0.86, 0.69 and 0.68). MiM-generated DVF errors, compared with the ground-truth DVFs, increase with increasing deformation intensity for all three sites. Certain sensitive structures (pituitary in HN and small bowel in pelvis) show poor deformation accuracy. DVF errors were not correlated with DSC or DTA (R: -0.649 to 0.450). After normalizing to organ volumes, DVF errors were strongly correlated with mean and max DTA (R: 0.939-0.945).
Conclusion: This study provided a comprehensive evaluation of intensity-based DIR algorithm using DVF errors and contour-based metrics. DIR accuracy is site and deformation dependent. The multiplication of DTA and organ volume can be used as a feasible indicator of DVF errors when ground-truth DVF is not accessible.
Funding Support, Disclosures, and Conflict of Interest: This work was supported by National Key Research and Development Program of China (2016YFC0103400). Jianfeng Q. was supported by the Taishan Scholars Program of Shandong Province (TS201712065).
Keywords
Image Processing, Validation, CT
Taxonomy
IM/TH- Image registration : CT
Contact Email
