Can Artificial Intelligence Beat Humans in Detecting Breast Malignancy on Mammograms?

Area of Science:

• Radiology
• Medical Imaging
• Oncology

Background:

• Mammography is crucial for early breast cancer detection.
• False-negative reports can delay diagnosis and treatment.
• Computer-Aided Detection (CAD) aims to improve mammographic accuracy.

Purpose of the Study:

• To evaluate the effectiveness of CAD in reducing false-negative mammography findings.
• To assess CAD's performance in detecting various breast cancer types and mammographic features.
• To compare CAD's diagnostic capabilities with human interpretation.

Main Methods:

• Retrospective analysis of 33 mammograms with biopsy-proven malignancy and 40 normal screening mammograms.
• Inclusion of lesion type, breast density, and CAD detection sensitivity.
• Correlation of mammographic findings with subsequent biopsy results.

Main Results:

• CAD achieved an overall sensitivity of 75.8% in detecting malignant lesions.
• High detection rates for masses (74%) and macrocalcifications (100%).
• Variable performance across different cancer types, with challenges in dense breasts and specific lesion morphologies.

Conclusions:

• CAD shows promise for detecting certain mammographic findings, particularly combined lesions and microcalcifications.
• CAD's reliability is limited in dense breasts, asymmetric densities, and with ill-defined lesions like invasive lobular carcinoma.
• Human interpretation remains essential for accurate breast cancer diagnosis, outperforming CAD in complex cases.