Encrypted login | home

Program Information

Validation of Deformable Image Registration Software Using a Standard CT Phantom

no image available
K Verdecchia

K Verdecchia*, L Qiu , A Magnelli , S Balik , S Kost , A Godley , P Xia , The Cleveland Clinic Foundation, Cleveland, OH


TU-RPM-GePD-JT-2 (Tuesday, August 1, 2017) 3:45 PM - 4:15 PM Room: Joint Imaging-Therapy ePoster Theater

Purpose: Image registration is frequently applied in radiotherapy clinics for adaptive replanning. Because of tumor anatomical changes and patient posture inconsistencies, deformable image registration (DIR) is often required. Prior to clinical use of DIR, validation of software algorithms is required but no standard approach exists currently. This study developed a simple procedure using a standard CT phantom to assess the accuracy of DIR.

Methods: One CT scan and two CBCT scans of a standard CT phantom (30 cm diameter) were acquired (3 mm slice thickness). For these images, three CT cylindrical inserts (3 cm diameter) that have electron density values similar to bone, lung and tissue, were locally deformed by a secondary software (ImSimQA). The CT phantom is readily available clinically and cylindrical inserts with well-defined electron densities minimize contouring uncertainties. To generate various deformed images with controlled severity, the length of the elongating vector was gradually incremented by 1 cm to 5 cm. The original and the deformed images were fused by a clinical DIR software algorithm (MIM version 6.6) and the dice similarity coefficient (DSC) quantified correlation to the original image.

Results: For all deformation severities (≤5 cm), the DSCs between the deformed CT and the original CT images were ≥ 0.984, 0.989 and 0.947 for bone, lung and tissue, respectively, demonstrating excellent shape recovery by deformation. The DSC values between the deformed CBCT and the original CBCTs ranged from 0.921-0.997 with an average of 0.968 ± 0.024. As expected due to inferior contrast, the poorest registration is observed by the tissue-like insert.

Conclusion: A simple procedure was developed to validate the performance of a software algorithm for DIR of CT and CBCT images. Further tests, such as the addition of noise and various slice thicknesses, will be conducted to further investigate the limitations of a DIR algorithm.

Funding Support, Disclosures, and Conflict of Interest: Partially supported by a grant from Elekta

Contact Email: