Validation of accuracy deformable image registration contour propagation using a benchmark virtual HN phantom dataset

Area of Science:

• Medical Imaging
• Radiotherapy Physics
• Computational Anatomy

Background:

• Virtual anatomic phantoms enable precise mapping of anatomical variations using deformation vector fields (DVFs).
• Dice Similarity Coefficient (DSC) and Mean Distance to Agreement (MDA) are standard metrics for evaluating deformable registration (DIR) algorithms.
• Head and Neck (HN) contour accuracy is critical for radiotherapy planning and adaptive monitoring.

Purpose of the Study:

• To validate the Accuray DIR algorithm for automatic contour propagation in HN virtual patient phantoms.
• To assess intermodal DIR accuracy between kilovoltage computed tomography (kVCT) and megavoltage computed tomography (MVCT) image data.
• To evaluate contour congruence using established metrics like DSC and MDA.

Main Methods:

• Utilized a HN virtual patient phantom dataset for kVCT-kVCT DIR validation with the Accuray DIR algorithm.
• Transformed kVCT image data to MVCT image quality to simulate adaptive monitoring scenarios.
• Quantified DIR accuracy using DSC and MDA, comparing results against ground truth contours and TG-132 recommendations.

Main Results:

• The Accuray DIR algorithm demonstrated adequate auto-propagation of HN contours, achieving average DSC of 0.8-0.9 and MDA within voxel width, aligning with TG-132 tolerances.
• Intermodal kVCT-MVCT DIR accuracy showed slightly reduced but acceptable agreement with ground truth contours.
• While generally accurate, contours critical for dosimetric planning require visual verification.

Conclusions:

• The Accuray DIR algorithm is suitable for routine auto-propagation of HN contours in radiotherapy.
• Standard MVCT image quality provides acceptable, though slightly less precise, results for DIR compared to kVCT.
• Clinical implementation should include mandatory visual inspection of critical contours to ensure dosimetric accuracy.