Segmentation of heterogeneous or small FDG PET positive tissue based on a 3D-locally adaptive random walk algorithm

Area of Science:

• Medical Imaging and Radiation Oncology
• Computational Biology and Bioinformatics

Background:

• Accurate tumor segmentation in Positron Emission Tomography (PET) is crucial for diagnosis and treatment planning.
• Existing segmentation methods struggle with small tumors or those exhibiting heterogeneous tracer uptake, such as Fluorodeoxyglucose (FDG).
• The Random Walk (RW) algorithm offers a promising approach but requires refinement for complex cases.

Purpose of the Study:

• To develop and evaluate an improved 3D segmentation algorithm, 3D-LARW, based on the Random Walk (RW) method.
• To enhance tumor delineation accuracy, particularly for small or heterogeneous FDG-avid lesions on PET images.
• To compare the performance of 3D-LARW against the original RW algorithm and other established segmentation techniques.

Main Methods:

• Developed the 3D-LARW algorithm, enhancing the original RW method with novel parameters.
• Incorporated Euclidean distance between voxels and probability densities of labels into the linear equations of the RW algorithm.
• Validated the algorithm using simulated heterogeneous spheres and clinical PET data from 14 patients, comparing against fixed thresholding, adaptive thresholding, and the FLAB method.

Main Results:

• The 3D-LARW algorithm demonstrated superior segmentation performance compared to the original RW method and the other evaluated techniques across simulated and clinical datasets.
• Significant improvements were observed in segmenting small tumors and those with heterogeneous FDG uptake.
• The enhanced parameters in 3D-LARW effectively addressed limitations of previous methods in complex tumor scenarios.

Conclusions:

• The 3D-LARW algorithm represents a significant advancement in PET image segmentation for oncological applications.
• This method provides more accurate delineation of challenging tumors, potentially improving treatment response assessment.
• 3D-LARW offers a robust tool for researchers and clinicians dealing with small or heterogeneous tumor burdens in PET imaging.