Voxel clustering for quantifying PET-based treatment response assessment.

Area of Science:

• Medical Imaging
• Radiotherapy Oncology
• Computational Biology

Background:

• Imaging biomarkers are vital for personalized cancer treatment assessment during therapy.
• Current methods use global measures, failing to capture heterogeneous tumor responses.
• There's a need for precise, local assessment of radiation treatment efficiency.

Purpose of the Study:

• To present an automated, multivoxel metric for assessing local radiation treatment efficiency.
• To analyze therapeutic response using multiple PET imaging studies over time.
• To move beyond global measures for a more granular understanding of tumor changes.

Main Methods:

• Employed level-set mathematics and deformable registration for PET image alignment.
• Classified voxels into response patterns (reduction/enhancement) using segmentation algorithms.
• Correlated signal enhancement patterns with radiation dose and anatomy for error analysis.

Main Results:

• Retrospectively applied the algorithm to PET/CT and radiotherapy data from 81 head and neck cancer cases (RTOG 0522 Trial).
• The technique accurately identified local metabolic changes, detecting segmentation misses.
• Results provided voxel-by-voxel analysis, detailed reports, and visual colorwash overlays.

Conclusions:

• The automated technique effectively analyzed treatment response in clinical cases.
• It serves as a valuable tool for accurate, outcome-based assessment of radiation therapy.
• The method is generalizable to other high-resolution diagnostic imaging modalities.