Quantifying the dose accumulation uncertainty after deformable image registration in head-and-neck radiotherapy

Area of Science:

• Medical Physics
• Radiotherapy
• Image Analysis

Background:

• Deformable image registration (DIR) is crucial for dose reconstruction and accumulation in radiotherapy.
• Assessing delivered dose and treatment plan validity during therapy is essential.
• Head-and-neck radiotherapy (HNRT) presents unique challenges due to anatomical changes.

Purpose of the Study:

• To investigate the uncertainty of reconstructed and accumulated dose in HNRT.
• To utilize in silico deformations based on clinical observations as ground truth.
• To assess structure-specific uncertainties of DIR algorithms.

Main Methods:

• Retrospective study on 12 HNRT patients using planning CT and cone beam CT data.
• Generation of in silico reference CBCTs and deformation vector fields (DVFs) using B-spline DIR.
• Assessment of inverse consistency (IC) and comparison of reconstructed dose for IC and non-IC voxels.

Main Results:

• 98.5% of voxels were inverse consistent.
• 95% confidence range for dose reconstruction: [-2.3%; +2.1%] for target structures, [-10.2%; +15.2%] for critical OARs, [-9.5%; +12.5%] for non-critical OARs.
• Inverse inconsistent voxels exhibited higher uncertainty.

Conclusions:

• Uncertainty in DIR-based accumulated dose can be quantified and integrated into dose-volume histograms (DVHs).
• This approach enables prospective, objective assessment of target coverage adequacy during treatment.
• Enhances treatment verification and plan adaptation in HNRT.