Semi-automatic integrated segmentation approaches and contour extraction applied to computed tomography scan images

Area of Science:

• Medical Imaging
• Computer Vision
• Radiology

Background:

• Traumatic brain injuries (TBIs) necessitate accurate segmentation on CT scans for diagnosis.
• Existing segmentation methods have limitations in precision and efficiency.

Purpose of the Study:

• To evaluate and compare various segmentation techniques for 2D CT scans of TBIs.
• To introduce a novel computational pipeline for enhanced lesion contour extraction.

Main Methods:

• Analysis of hybrid, feature extraction, level sets, region growing, and watershed segmentation methods.
• Validation using pixel intensities, gradient magnitude, affinity maps, and catchment basins.
• Development of a new pipeline employing bilateral filtering, diffusion, watershed, and mathematical morphology for contour extraction.

Main Results:

• Parametric and nonparametric arguments of segmentation methods were analyzed.
• The new computational pipeline demonstrated effectiveness in contour extraction based on gradient functions.
• Lesion classification evaluations are ongoing and detailed in separate work.

Conclusions:

• Multiple segmentation methods show potential for TBI analysis on CT scans.
• The developed computational pipeline offers a promising approach for accurate lesion boundary detection.
• Further research in pattern recognition is needed for comprehensive lesion classification.