Sensitivity study of voxel-based PET image comparison to image registration algorithms

Area of Science:

• Medical imaging analysis
• Radiotherapy and oncology
• Image registration techniques

Background:

• Accurate voxel-based comparison of sequential PET images is vital for treatment adaptation and response assessment.
• The choice of deformable registration algorithm can influence treatment evaluation outcomes.
• This study aimed to assess the impact of different registration algorithms on therapy response assessment.

Purpose of the Study:

• Investigate the influence of various deformable registration algorithms on the evaluation of lung tumor response to chemoradiotherapy.
• Compare the performance of rigid versus multiple deformable registration methods in assessing treatment response.
• Determine the impact of tumor deformation on the accuracy of response assessment.

Main Methods:

• 16 patients with 20 lung tumors underwent pre- and post-treatment CT and 4D FDG-PET scans.
• CT images were registered using rigid and ten deformable algorithms; transformations were applied to PET images.
• Tumor subvolumes (progressor, stable-disease, responder) were classified based on SUV changes; agreement was assessed using Dice Similarity Index (DSI).

Main Results:

• Median DSI varied across subvolumes and algorithms, with substantial variability (CV > 10%) for both rigid and deformable methods.
• Tumor deformation significantly impacted DSI for progressor (R = -0.46 to -0.60) and stable-disease (R = -0.71 to -0.72) subvolumes.
• Deformable registration was not essential for responder subvolumes (insignificant |R| < 0.26) or when tumor deformation was low (RMSrigid < 150 HU).

Conclusions:

• The selection of deformable registration algorithms significantly affects voxel-based PET image comparisons and treatment response evaluation.
• For cases with minimal tumor deformation (RMSrigid < 150 HU), rigid and deformable registration methods produce comparable results.
• Deformable registration is not always necessary, particularly when tumor deformation is low, simplifying treatment response assessment.