Active contour algorithm with discriminant analysis for delineating tumors in positron emission tomography

Area of Science:

• Medical Imaging
• Computational Biology
• Radiology

Background:

• Accurate cancer delineation in positron emission tomography (PET) datasets is crucial for effective treatment planning and response assessment.
• Current segmentation methods often require significant user input, leading to variability and inefficiency.
• Developing automated, real-time 3D segmentation techniques is essential for clinical workflow optimization.

Purpose of the Study:

• To develop and validate a fully automated, real-time 3D segmentation algorithm for cancer delineation in PET imaging.
• To minimize user dependency and improve the objectivity and reproducibility of tumor segmentation.
• To assess the algorithm's performance using both phantom and clinical datasets with various radio-tracers and body districts.

Main Methods:

• A novel approach combining a preliminary user-guided initialization with an automated region of interest identification based on the highest standardized uptake value (SUV).
• Three-dimensional volume reconstruction using a slice-by-slice marching method with an automatic stop condition.
• Enhanced local active contour segmentation on each slice, integrating discriminant analysis (a machine learning component) into a new energy functional.
• Validation using phantom experiments (spheres, zeolites) and clinical cases (lung, brain, head and neck) with 18F-fluoro-2-deoxy-d-glucose (FDG) and 11C-methionine.

Main Results:

• Phantom studies demonstrated high accuracy with Dice Similarity Coefficients above 90% for spheres and 80% for zeolites.
• Clinical case analysis showed excellent agreement with the gold standard, achieving an R² value of 0.98.
• The algorithm proved to be almost completely automatic, producing user-independent segmentation results.

Conclusions:

• The proposed automated 3D segmentation method is highly efficient and accurate for cancer delineation in PET imaging.
• This technique has the potential to be integrated into routine clinical practice, aiding in treatment response assessment and radiotherapy targeting.
• The user-independent nature and high performance suggest a significant benefit for oncological imaging workflows.