Atlas selection methods for multi-atlas-based segmentation in breast cancer radiotherapy planning.

Area of Science:

• Medical Imaging
• Radiotherapy Planning
• Computational Anatomy

Background:

• Accurate segmentation of breast cancer is crucial for effective radiotherapy planning.
• Multi-atlas-based segmentation (MABS) is a promising technique, but atlas selection impacts performance.
• Optimizing atlas selection is key to improving MABS efficiency and accuracy.

Purpose of the Study:

• To evaluate different atlas selection strategies for MABS in breast cancer radiotherapy.
• To assess the impact of patient stratification on segmentation accuracy and computational time.
• To identify optimal atlas selection criteria for reliable breast cancer segmentation.

Main Methods:

• Developed and applied MABS using 30 patient atlases and 15 test cases.
• Stratified atlases into groups based on breast separation, height, and volume.
• Calculated Dice similarity coefficient (DSC) to compare MABS with manual segmentation.

Main Results:

• Increased atlas selection led to higher Dice similarity coefficients (DSC).
• Patient stratification significantly reduced computational time compared to using all atlases.
• Height-matched and volume-matched atlas selection yielded median DSC values greater than 0.9.

Conclusions:

• Atlas selection is critical for optimizing MABS in breast cancer radiotherapy.
• Patient stratification, particularly using breast height, offers an efficient method for atlas selection.
• Breast height may serve as a practical predictor for breast volume in MABS.