Influence of filter choice on 18F-FDG PET segmentation accuracy determined using generalized estimating equations

Area of Science:

• Nuclear Medicine Imaging
• Medical Image Analysis
• Biomedical Engineering

Background:

• Accurate segmentation of fluoro-deoxy-glucose (FDG)-positron emission tomography (PET) images is crucial for quantitative analysis.
• Image filtering is a common preprocessing step that can influence segmentation performance.
• Traditional statistical methods may not fully account for the inherent properties of segmentation quality metrics.

Purpose of the Study:

• To evaluate the impact of different image filters on various FDG-PET segmentation techniques.
• To introduce and validate the use of generalized estimating equations (GEEs) for analyzing segmentation quality.
• To provide a more statistically rigorous approach for assessing segmentation performance in nuclear medicine.

Main Methods:

• Simulated PET images with spherical and irregular objects were generated at varying contrasts and imaging durations.
• Images were processed using Gaussian and bilateral filters with different full-width half maximum (FWHM) values, with and without pre-smoothing.
• Four segmentation methods (thresholding, adaptive thresholding, k-means, region-growing) were applied and evaluated using Dice similarity coefficient (DSC) and symmetric-mean-absolute-surface-distance (SMASD).
• GEE models were fitted to account for the correlation structure of DSC values.

Main Results:

• All segmentation methods achieved mean DSC values between 0.71-0.87 and mean SMASD values between 0.72-2.10 mm across tested filters.
• The bilateral filter with 3 mm Gaussian pre-smoothing showed comparable performance to a 5 mm Gaussian filter (DSC 0.80 vs. 0.79).
• GEE analysis revealed correlations of 0.118 for spheres and 0.290 for irregular objects, demonstrating the model's ability to capture dependencies.

Conclusions:

• Filter selection critically influences FDG-PET segmentation outcomes, with bilateral and Gaussian filters showing similar efficacy under specific parameters.
• GEE provides a statistically sound framework for analyzing segmentation quality metrics, offering more accurate estimations than simpler methods.
• This approach enhances the reliability of experimental effect estimations in nuclear medicine segmentation studies.