Accuracy of target delineation by positron emission tomography-based auto-segmentation methods after deformable image...

Area of Science:

• Medical imaging
• Radiology
• Nuclear medicine

Background:

• Diagnostic PET/CT images are fused with planning CT images using deformable image registration (DIR).
• Evaluating the impact of DIR on quantitative PET metrics and target delineation is crucial for accurate treatment planning.

Purpose of the Study:

• To assess the accuracy of standardized uptake value (SUV) measurements after DIR.
• To evaluate the effect of DIR on target delineation using various auto-segmentation methods on PET images.

Main Methods:

• A phantom study using spheres and ellipsoids filled with 18F-FDG was conducted.
• PET/CT images were acquired, and auto-segmentation (SUV2.5, SUV40%max, gradient-based) was performed on original and deformed images after DIR.
• Volumes and SUVmax were compared between original and deformed PET images.

Main Results:

• DIR caused a minor decrease in SUVmax (mean absolute difference: -0.10 ± 0.04).
• SUV2.5 and SUV40%max methods showed volume differences <5% after DIR.
• Gradient-based segmentation resulted in larger contoured volumes on deformed images, with a maximum volume difference of -22.8% when deformed to a smaller sphere.

Conclusions:

• DIR has a negligible effect on SUV measurements in PET/CT.
• Gradient-based auto-segmentation for target delineation on deformed PET images requires caution due to potential intensity profile distortions and significant volume changes.