Variational PET/CT Tumor Co-segmentation Integrated with PET Restoration

Area of Science:

• Medical Imaging
• Computational Biology
• Radiology

Background:

• Positron Emission Tomography (PET) and Computed Tomography (CT) are crucial in radiation oncology.
• PET offers high contrast but blurry tumor edges; CT provides high resolution but low soft-tissue contrast.
• Tumor segmentation from single PET or CT images is challenging due to complementary information limitations.

Purpose of the Study:

• To develop a novel variational method for accurate tumor co-segmentation in PET/CT scans.
• To address the challenge of inconsistent information between PET and CT for precise tumor edge localization.
• To improve segmentation accuracy by adaptively fusing complementary data from PET and CT.

Main Methods:

• Proposed a variational method based on the Gamma-convergence approximation of the Mumford-Shah segmentation model.
• Implemented a fusion strategy to adaptively handle inconsistent information between PET and CT.
• Integrated a PET restoration process to mitigate uncertainties from blurry tumor edges.

Main Results:

• The method achieved high accuracy in PET/CT co-segmentation and PET restoration.
• Successfully estimated the blur kernel of the PET scanner.
• Demonstrated accurate tumor segmentation even for complex cases with inhomogeneous Fluorodeoxyglucose (FDG) uptake or adjacent tissue invasion.
• Achieved an average Dice Similarity Index (DSI) of 0.85 ± 0.06, Volume Error (VE) of 0.09 ± 0.08, and Classification Error (CE) of 0.31 ± 0.13.

Conclusions:

• The proposed adaptive fusion method effectively overcomes information inconsistency in PET/CT co-segmentation.
• The integrated PET restoration enhances segmentation accuracy and PET scanner characterization.
• This approach offers a robust solution for precise tumor segmentation in radiation oncology, improving treatment planning.