Program Information
Evaluation of Image Registration Accuracy for Tumor and OARs in the Thorax for Compliance with TG 132 Recommendations
C Guy*, E Weiss , S Che , W Sleeman IV , M Rosu , Virginia Commonwealth University, Richmond, VA
Presentations
SU-I-GPD-J-66 (Sunday, July 30, 2017) 3:00 PM - 6:00 PM Room: Exhibit Hall
Purpose: To evaluate registration accuracy for in-house (BSplines) and MIM (free form) deformable registration methods and assess the appropriateness of registration accuracy recommendations by TG 132. The deformations investigated encompassed respiratory motion and change in patient orientation (Supine-Inhale SI, Supine-Exhale SE, Prone-Inhale PI, Prone-Exhale PE).
Methods: 4DCTs for 12 lung cancer patients were acquired in prone and supine positions. Tumor and OARs (esophagus, lungs + lobes, heart, T2, T5, and T12) were delineated by a physician on all datasets. SI was registered to SE, PI, and PE using both registration methods. All contours from SI were propagated using the resulting transformations and compared to physician-delineated contours using dice similarity coefficient (DSC), mean distance to agreement (MDA), and Hausdorff distance (HD). The Jacobian determinant was used to evaluate the registration quality inside the structures.
Results: Across all patients, the mean(stdev) of metrics were: DSC 0.8(0.1), MDA 2.1(1.4) mm, HD 22.6(17.7) mm for MIM, and DSC 0.8(0.2), MDA 2.3(1.4) mm, HD 22.5(17.2) mm for in-house methods. MIM performed better with complex/low contrast structures (esophagus, vertebrae), while in-house method performed better with lungs (whole and individual lobes). Accuracy metrics worsened for both methods when propagating from supine to prone, but most structures remained within tolerances recommended by TG 132 and other literature. Esophagus often fell out of tolerance across all registrations, likely due to poor contrast and high delineation uncertainty of the structure. Most contours met accuracy constraints (DSC > 0.8, MDA < 3 mm); however, the Jacobian determinant contained regions with negative values, indicating non-physical deformation (folding of space).
Conclusion: Most contour-based accuracy limits were met for thoracic structures, but irregularities and negative values were observed in the Jacobian determinants. Therefore, the registration methods tested were sufficiently accurate for contour propagation, but not for more complex uses such as dose accumulation.
Funding Support, Disclosures, and Conflict of Interest: Elisabeth Weiss receives UpToDate royalties and has NIH grants, a Philips research agreement, and a Varian licensing agreement.
Contact Email: