Using patient-specific phantoms to evaluate deformable image registration algorithms for adaptive radiation therapy

Area of Science:

• Medical Physics
• Radiotherapy
• Image Analysis

Background:

• Adaptive treatment planning in radiotherapy relies on accurate deformable image registration (DIR).
• Evaluating DIR algorithm performance is essential for optimizing adaptive radiotherapy strategies.

Purpose of the Study:

• To assess the performance of B-spline-based deformable multipass (DMP) and deformable demons (Demons) DIR algorithms.
• To evaluate these algorithms using both computational and physical phantoms.

Main Methods:

• Developed 11 computational models using finite element method (FEM) from patient CT scans.
• Created lung and prostate image phantoms with FEM-generated displacement vector fields (DVF).
• Incorporated image noise into prostate phantoms to simulate CBCT and evaluated algorithms on a physical lung phantom.

Main Results:

• DMP algorithm showed mean errors of 1.0–3.1 mm (computational) and 1.9 mm (physical lung phantom).
• Prostate phantom errors ranged from 1.0–2.4 mm, with higher errors in regions of large displacement.
• Observed sinusoidal errors in DMP registrations and patient-dependent accuracy.

Conclusions:

• DIR algorithm accuracy is influenced by patient-specific factors like deformation magnitude and image gradients.
• Individual verification of DIR algorithms is recommended for adaptive radiation therapy implementation.