AMS-U-Net: automatic mass segmentation in digital breast tomosynthesis via U-Net

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Digital Breast Tomosynthesis (DBT) is crucial for breast cancer screening.
• Increasing DBT slice counts elevate mass contouring workload and reduce efficiency.
• Automated segmentation methods are needed to address these challenges.

Purpose of the Study:

• To develop AMS-U-Net, a fully automatic mass segmentation method for DBT.
• To improve efficiency and accuracy in breast cancer screening using DBT.

Main Methods:

• The study employed a four-stage approach: pre-processing, AMS-U-Net training, segmentation, and post-processing.
• Performance was evaluated on 50 DBT slices using metrics like True Positive Ratio (TPR), False Positive Ratio (FPR), F-score, Intersection over Union (IoU), and 95% Hausdorff distance.
• The model was designed to handle class imbalance inherent in medical imaging datasets.

Main Results:

• The AMS-U-Net model achieved a TPR of 0.911, FPR of 0.003, F-score of 0.911, IoU of 0.900, and a 95% Hausdorff distance of 5.82 pixels.
• These quantitative results indicate high accuracy in mass segmentation.

Conclusions:

• AMS-U-Net demonstrates high accuracy in segmenting breast masses from DBT images without manual intervention.
• The automated approach shows significant potential to enhance clinical efficiency and workflow in breast cancer screening.
• This technology can streamline the interpretation of DBT scans, leading to faster diagnoses.