A gradient-based method for segmenting FDG-PET images: methodology and validation.

Area of Science:

• Medical Imaging
• Image Segmentation
• Nuclear Medicine

Background:

• Accurate segmentation of FDG-PET images is crucial for precise tumor volume delineation.
• Existing methods, like thresholding, can lead to significant overestimation of tumor volumes.
• Image noise and blurring in PET scans can compromise segmentation accuracy.

Purpose of the Study:

• To introduce and validate a new gradient-based image segmentation method for FDG-PET scans.
• To compare the accuracy of the proposed gradient-based method against a threshold-based approach.
• To assess the impact of image denoising and deblurring on segmentation performance.

Main Methods:

• The proposed method employs watershed transform and hierarchical cluster analysis.
• Iteratively reconstructed PET images were preprocessed using edge-preserving filters and constrained iterative deconvolution.
• Validation involved computer-generated phantoms, a physical phantom, and PET images from seven laryngeal cancer patients.

Main Results:

• Image deconvolution reduced volume and radius mis-estimates in phantoms.
• The gradient-based method showed slight underestimation (10-20%) in phantom sphere volumes but negligible radius differences.
• For laryngeal tumors, the gradient-based method's volumes closely matched macroscopic specimens, unlike the threshold-based method which overestimated by 68% (p=0.014).

Conclusions:

• The gradient-based segmentation method, applied to preprocessed FDG-PET images, demonstrates superior accuracy over the source-to-background ratio thresholding method.
• This gradient-based approach offers a more reliable tool for delineating tumor volumes in PET imaging.
• Further refinement may be needed as neither method fully encompassed macroscopic specimens.