Unsupervised feature correlation model to predict breast abnormal variation maps in longitudinal mammograms.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Breast cancer remains a leading cause of mortality in women worldwide.
• Early detection and accurate diagnosis are crucial for improving patient prognosis and survival rates.
• Traditional screening methods have limitations in detecting subtle abnormalities.

Purpose of the Study:

• To develop a novel unsupervised feature correlation network for enhanced early breast cancer detection.
• To improve the accuracy and reliability of diagnosing breast abnormalities using longitudinal 2D mammograms.
• To reduce breast cancer mortality through more precise diagnostic capabilities.

Main Methods:

• A novel unsupervised feature correlation network was developed.
• The model utilizes longitudinal 2D mammograms (current and prior year) for comparative analysis.
• Key components include a feature correlation module, attention suppression gate, and abnormality detection module.

Main Results:

• The proposed model generates breast abnormal variation maps.
• It accurately distinguishes between normal and cancerous mammograms, outperforming baseline models.
• Significant improvements were observed in Accuracy, Sensitivity, Specificity, Dice score, and cancer detection rate.

Conclusions:

• The developed unsupervised network offers a promising advancement in breast cancer early detection.
• This approach enhances diagnostic accuracy by analyzing longitudinal mammographic data.
• The model has the potential to significantly improve patient outcomes and reduce breast cancer mortality.