Efficient selection of the most similar image in a database for critical structures segmentation

Area of Science:

• Medical Imaging
• Radiotherapy Planning
• Computational Anatomy

Background:

• Accurate delineation of critical structures is essential for effective radiotherapy planning.
• Atlas-based segmentation is efficient for brain structures but challenging for the head and neck due to image variability and potential over-segmentation.
• Existing methods often require extensive computational resources for atlas construction.

Purpose of the Study:

• To develop a computationally efficient method for selecting the most similar template image for radiotherapy segmentation.
• To improve the specificity of segmentation in the head and neck region compared to traditional atlas-based methods.
• To address the limitations of atlas construction in highly variable anatomical regions.

Main Methods:

• A novel approach for template selection based on the distance between transformations.
• Registration of patient images to an average image, rather than every sample in the database, for template matching.
• Leave-One-Out cross-validation on a dataset of 45 patients for evaluation.

Main Results:

• The proposed method demonstrates significant improvement in the specificity of segmented structures.
• Qualitative and quantitative comparisons show superior performance over classical atlas-based segmentation.
• The approach is computationally more efficient than traditional methods.

Conclusions:

• The developed template selection method offers a more efficient and accurate alternative for radiotherapy planning, particularly in the head and neck region.
• This technique enhances segmentation specificity, crucial for precise radiation delivery.
• The method's computational efficiency makes it a valuable tool for clinical applications.