Objective PET lesion segmentation using a spherical mean shift algorithm

Area of Science:

• Oncology
• Medical Imaging
• Image Analysis

Background:

• Positron Emission Tomography (PET) imaging is vital in oncology for lesion characterization and therapy response assessment.
• Accurate lesion delineation is challenging due to factors like small tumor size, blurred boundaries, motion, and inhomogeneous uptake, leading to clinical variability.
• Existing methods often lack objectivity, introducing operator variability in lesion assessment.

Purpose of the Study:

• To develop and analyze an objective segmentation algorithm for PET lesions.
• To address the challenges of accurate lesion delineation in PET imaging.
• To reduce operator variability in clinical assessments.

Main Methods:

• A novel segmentation algorithm based on the mean shift algorithm is proposed.
• The algorithm is applied in a spherical coordinate frame for directional assessment.
• A varying background model is incorporated to improve segmentation accuracy.

Main Results:

• The algorithm yields objective segmentations, minimizing operator variability.
• Analysis using clinically relevant hybrid digital phantoms demonstrates effectiveness.
• The technique shows improved performance relative to other existing methods.

Conclusions:

• The developed PET lesion segmentation algorithm offers objective and reproducible assessments.
• This technique has the potential to enhance the reliability of oncology imaging interpretation.
• Improved lesion delineation can lead to more accurate therapy response evaluation in cancer patients.