A feature-based approach for atlas selection in automatic pelvic segmentation.

Area of Science:

• Medical Image Analysis
• Computational Anatomy
• Radiotherapy Planning

Background:

• Accurate automatic segmentation is crucial for clinical applications, including radiotherapy.
• Existing atlas-based segmentation methods face limitations due to insufficient atlas data and computational constraints.

Purpose of the Study:

• To propose and evaluate a novel atlas selection procedure (MAS-SAGA) for enhanced multi-atlas-based segmentation.
• To compare the performance of feature-based (MAS-FASA) and similarity-based (MAS-SIM) atlas selection methods.

Main Methods:

• Developed the MAS-SAGA approach, integrating image similarity and volume features for atlas selection.
• Utilized a dataset of anonymized female pelvic CT images for segmentation of bladder and rectum.
• Employed a three-fold cross-validation strategy to assess segmentation accuracy and computational efficiency.

Main Results:

• MAS-SAGA demonstrated superior performance over conventional multi-atlas-based segmentation (cMAS) in Dice Similarity Coefficient (DSC) and 95th Percentile Hausdorff Distance (95HD) for bladder and rectum.
• The proposed method significantly reduced computation time compared to cMAS.
• MAS-FASA identified different atlases than MAS-SIM, leading to overall improved segmentation outcomes.

Conclusions:

• The MAS-SAGA procedure offers a promising advancement for medical image segmentation, enhancing accuracy and efficiency.
• Feature-based atlas selection techniques show potential for improving the efficacy of multi-atlas segmentation algorithms.