A YOLO-based AI system for classifying calcifications on spot magnification mammograms.

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Distinguishing malignant from benign calcifications on mammograms is crucial for patient management.
• Unnecessary biopsies lead to increased healthcare costs and patient anxiety.

Purpose of the Study:

• To investigate the efficacy of a deep learning-based AI system in classifying calcifications on spot magnification mammograms.
• To assess the potential of the AI system to reduce unnecessary biopsies.

Main Methods:

• A retrospective study utilizing public (CBIS-DDSM) and in-house mammogram datasets.
• Development of an AI system using a You Only Look Once (YOLO) based adaptive multiscale decision fusion module.
• Performance evaluation via receiver operating characteristic (ROC) analysis.

Main Results:

• The AI system achieved an area under the ROC curve of 0.888 on the in-house dataset.
• Sensitivity was 88.4%, specificity 80.8%, and accuracy 84.6% at the optimal cutoff.
• The system could potentially avoid 80.8% of benign biopsies when used with two views of spot magnification mammograms.

Conclusions:

• The AI system demonstrates strong accuracy in classifying calcifications on mammograms.
• This technology has the potential to significantly decrease the rate of unnecessary biopsies for suspicious calcifications.