WE-E-213CD-08: A Novel Level Set Active Contour Algorithm Using the Jensen-Renyi Divergence for Tumor Segmentation in...

Area of Science:

• Medical Imaging
• Radiotherapy
• Computational Biology

Background:

• Accurate delineation of biological tumor volume (BTV) is crucial for image-guided radiotherapy.
• Positron Emission Tomography (PET) offers valuable insights but faces segmentation challenges due to low resolution and noise.
• Traditional active contour methods struggle with noise and low-contrast regions in PET imaging.

Purpose of the Study:

• To evaluate a novel active contour algorithm for enhanced PET tumor segmentation.
• To address the limitations of existing methods in segmenting noisy and low-contrast PET data.
• To assess the performance of a Jensen-Renyi Divergence-based active contour model for BTV delineation.

Main Methods:

• A novel active contour segmentation algorithm maximizing Jensen-Renyi Divergence was applied to PET scans of 7 patients.
• The algorithm was implemented using GPU acceleration for efficient processing.
• PET images were pre-processed with denoising and deconvolution techniques.
• Segmentation results were compared against histology-derived macroscopic BTV contours from the Louvain database.

Main Results:

• The algorithm achieved good agreement with macroscopic contours, with a concordance index of 0.6 ± 0.09.
• A classification error of 55 ± 16.5% was observed.
• Each iteration of the algorithm took 0.5-1.3 seconds, reaching convergence in 10-30 iterations.

Conclusions:

• The Jensen-Renyi active contour method demonstrates competitive and often superior performance compared to other PET segmentation methods.
• The algorithm shows promise for improving PET tumor segmentation accuracy in radiotherapy.
• Further validation on larger datasets and performance optimization are recommended for clinical application.