An unsupervised automatic segmentation algorithm for breast tissue classification of dedicated breast computed...

Area of Science:

• Medical Imaging
• Radiology
• Computer-Aided Diagnosis

Background:

• Accurate breast tissue classification is crucial for diagnostic imaging and research.
• Dedicated breast CT (bCT) offers detailed anatomical information but requires robust segmentation methods.
• Existing methods may lack accuracy or robustness in complex breast tissue environments.

Purpose of the Study:

• To develop and evaluate an automated algorithm for classifying skin, vasculature, adipose, and fibroglandular tissues in bCT images.
• To assess the algorithm's accuracy, robustness to image noise, and performance compared to prior approaches.

Main Methods:

• Combined intensity- and region-based segmentation with active contours and data mining.
• Utilized region-growing for skin, active contours for adipose tissue, and k-means clustering for vasculature.
• Validated against manual segmentation (gold standard) using Dice similarity coefficient (DSC) and Hausdorff distance.

Main Results:

• Achieved high average DSC (90-95%) across different bCT systems and datasets.
• Demonstrated robustness to image noise, maintaining high accuracy at increasing noise levels (71-85%).
• Outperformed previous methods with a global average DSC of 94.5% compared to 87%.

Conclusions:

• The developed algorithm provides accurate and robust automated breast tissue classification without prior training.
• Potential applications include breast density quantification, cancer biomarker characterization, and radiation dose analysis.
• Facilitates further research in breast imaging through improved segmentation and phantom design.