Joint two-view information for computerized detection of microcalcifications on mammograms

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Breast Cancer Screening

Background:

• Accurate detection of microcalcification clusters in mammograms is crucial for early breast cancer diagnosis.
• Current single-view detection methods often struggle with high false-positive rates.
• Utilizing joint information from multiple mammographic views can potentially enhance detection accuracy.

Purpose of the Study:

• To develop and evaluate a novel computerized technique for microcalcification detection using joint two-view information.
• To improve the accuracy of microcalcification cluster detection by reducing false positives.
• To assess the performance of a fusion method combining single-view and two-view classifiers.

Main Methods:

• A two-view detection algorithm was developed, pairing candidates from craniocaudal (CC) and mediolateral-oblique (MLO) views based on radial distance from the nipple.
• Candidate pairs were classified using a similarity classifier leveraging joint information, alongside a single-view classifier.
• A fusion method combined the outputs of both classifiers to distinguish true microcalcification clusters from false positives (FP).
• The algorithm was trained on the University of South Florida (USF) database and tested on an independent University of Michigan (UM) dataset.

Main Results:

• The similarity classifier achieved a low FP rate but had limited sensitivity (69%).
• The single-view classifier had a higher FP rate but reached higher sensitivity (93%).
• The fusion method substantially reduced FPs at medium sensitivities while maintaining high maximum sensitivity.
• At 80% sensitivity for malignant clusters, the two-view fusion method yielded an FP rate of 0.18, compared to 0.35 for the single-view method on the UM test set.

Conclusions:

• Correspondence of microcalcification cluster candidates across two different mammographic views provides valuable information for improving detection accuracy.
• The developed two-view fusion technique offers a promising approach to reduce false positives in computerized microcalcification detection.
• This method has the potential to enhance the efficiency and reliability of mammographic screening for breast cancer.