Segmentation method comparison for baseline [¹⁸F]FDG PET-CT in follicular lymphoma patients.

Area of Science:

• Nuclear Medicine
• Oncology
• Medical Imaging Analysis

Background:

• Follicular lymphoma (FL) is a common non-Hodgkin lymphoma.
• ¹⁸F]FDG PET-CT is crucial for FL management, but accurate tumor delineation is vital for prognostic assessment.
• Standardized segmentation methods for FL PET are needed due to varying uptake patterns.

Purpose of the Study:

• To evaluate single-threshold and multi-threshold segmentation methods for FL [¹⁸F]FDG PET-CT.
• To assess segmentation quality, interobserver variability, and ease-of-use of different methods.

Main Methods:

• Two observers segmented baseline PET-CT scans from 37 FL patients using 13 semi-automatic methods (6 single-threshold, 7 multi-threshold).
• Methods included SUV threshold, AI-based, majority vote, and lesion-based approaches.
• Segmentation quality, interobserver volume differences, and manual adjustment effort were quantified.

Main Results:

• All tested methods showed generally satisfactory performance.
• Multi-threshold methods achieved better segmentation quality (less under/oversegmentation) than single-threshold methods.
• The SUV4.0 single-threshold method exhibited low interobserver variability (median 0.3 mL IQR 2.0 mL).

Conclusions:

• SUV4.0 is recommended among single-threshold methods for its ease-of-use, low variability, and acceptable quality.
• Multi-threshold lesion-based methods offer superior segmentation quality but are more complex.
• SUV4.0 and a lesion-based method are identified as candidates for further clinical evaluation in FL.