Improving the Quantitative Analysis of Breast Microcalcifications: A Multiscale Approach

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Breast Cancer Diagnostics

Background:

• Accurate characterization of microcalcifications (MCs) in 2D digital mammography is crucial for reducing diagnostic uncertainty and callbacks.
• Automated identification and segmentation of MCs are challenging, often resulting in high false positive rates.

Purpose of the Study:

• To develop and validate a two-stage multiscale approach for MC segmentation in full-field digital mammograms (FFDMs) and magnification views.
• To improve the accuracy of MC segmentation and enhance the classification of benign versus malignant calcifications.

Main Methods:

• A two-stage approach combining blob detection, Hessian analysis, and a regression convolutional network for MC segmentation.
• Training and validation on 435 screening and diagnostic FFDMs from two datasets.
• Feature extraction and classification using gradient tree boosting on magnification views of amorphous MCs.

Main Results:

• Superior mean intersection over the union (0.670 ± 0.121) compared to state-of-the-art methods (0.524 ± 0.034).
• Achieved a true positive rate of 0.744 versus 0.581 at 0.4 false positive detections per square centimeter.
• Outperformed comparison methods in distinguishing benign from malignant amorphous calcifications with an AUC of 0.763 versus 0.710.

Conclusions:

• The proposed two-stage multiscale approach significantly improves MC segmentation accuracy in digital mammography.
• This method enhances the ability to differentiate between benign and malignant microcalcifications, aiding in breast cancer diagnosis.