A feature-based approach for refinement of model-based segmentation of low contrast structures

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Automated medical image segmentation requires accuracy and robustness.
• Model-based segmentation (MBS) uses prior shape information but struggles with low-contrast images, such as soft tissues in CT scans.
• Existing methods face challenges in feature response generation for difficult-to-segment tissues.

Purpose of the Study:

• To enhance a framework for voxel classification-based refinement of MBS.
• To improve segmentation accuracy for low-contrast medical images.
• To introduce a novel feature weighting methodology for better classifier performance.

Main Methods:

• Utilized a level-set segmentation technique with shape priors.
• Implemented a voxel classification-based refinement of model-based segmentation.
• Developed and applied a novel feature weighting methodology for classifier improvement.

Main Results:

• Achieved superior performance compared to previous feature selection methods.
• Demonstrated fully automated segmentation of low-contrast organs in head and neck CT data.
• Reported an increase of up to 22% in segmentation accuracy compared to the previous approach.

Conclusions:

• The enhanced framework significantly improves automated segmentation of challenging medical images.
• The novel feature weighting methodology is effective in boosting classifier performance.
• This work offers a more accurate solution for segmenting low-contrast soft tissue organs in CT scans.