MRI Brain Tumor Segmentation and Necrosis Detection Using Adaptive Sobolev Snakes

Area of Science:

• Medical Imaging
• Neuroscience
• Computer Vision

Background:

• Brain tumor segmentation is crucial for neurosurgical planning and disease staging.
• Accurate assessment of tumor shape and necrosis over time is vital for evaluating disease progression.
• Existing methods may face challenges with low-contrast imagery and noise.

Purpose of the Study:

• To develop a semi-automatic algorithm for precise brain tumor segmentation and necrosis detection in MRI volumes.
• To improve the evaluation of tumor characteristics for disease progression monitoring.
• To provide a robust segmentation method suitable for clinical applications.

Main Methods:

• A novel algorithm converting MRI volumes into a probability space using an online learned model.
• Tumor probability density estimation via anisotropic 3D diffusion.
• Adaptive segmentation using the Sobolev active contour (snake) algorithm.
• Necrosis detection by identifying outliers in the probability distribution within segmented regions.

Main Results:

• The algorithm demonstrated robustness to noise and insensitivity to manual initialization.
• Effective segmentation was achieved even in low-contrast MRI imagery.
• Necrosis regions were detected, with small, potentially noisy regions being excluded.
• Segmentation results closely matched expert manual segmentations.

Conclusions:

• The proposed semi-automatic algorithm offers accurate and robust brain tumor segmentation and necrosis detection.
• This method enhances the ability to monitor disease progression through detailed tumor analysis.
• The algorithm shows promise for clinical utility in neuro-oncology and neurosurgery.