Does consensus contours improve robustness and accuracy on F-FDG PET imaging tumor delineation?

Area of Science:

• Medical Imaging
• Radiology
• Oncology

Background:

• Accurate tumor segmentation in 2-deoxy-2-[F]fluoro-D-glucose (F-FDG) PET imaging is crucial for nasopharyngeal carcinoma (NPC) treatment planning.
• Variability in segmentation methods can impact treatment efficacy and reproducibility.

Purpose of the Study:

• To evaluate the robustness and accuracy of consensus contours (ConSeg) for NPC tumor segmentation using F-FDG PET imaging.
• To compare ConSeg performance against other automatic segmentation methods.

Main Methods:

• Segmentation was performed on 225 NPC F-FDG PET datasets and 13 extended cardio-torso (XCAT) simulations using active contour, affinity propagation (AP), contrast-oriented thresholding (ST), and 41% maximum tumor value (41MAX).
• Consensus contours (ConSeg) were generated using a majority vote rule.
• Metabolically active tumor volume (MATV), relative error (RE), and Dice similarity coefficient (DSC) were used for quantitative analysis, along with test-retest (TRT) metrics.

Main Results:

• Affinity propagation (AP) showed the highest variability in MATV.
• ConSeg demonstrated better TRT performance for MATV compared to AP, but was slightly poorer than ST or 41MAX.
• Average segmentation (AveSeg) showed comparable or better accuracy than ConSeg.
• Irregular initial masks improved RE and DSC for AP, AveSeg, and ConSeg compared to rectangular masks.
• All methods underestimated tumor boundaries in XCAT simulations with respiratory motion.

Conclusions:

• Consensus contours provide a robust approach to reduce segmentation variability in F-FDG PET imaging for NPC.
• The consensus method did not significantly improve segmentation accuracy on average.
• Irregular initial masks may contribute to mitigating segmentation variability.