Related Experiment Video
Updated: Jun 14, 2025
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
Exploring the Efficacy of Artificial Intelligence-Based Computer-Aided Detection for Breast Cancer Detection on
Purpose:
In this retrospective study, we aimed to assess the efficacy of artificial intelligence-based computer-aided detection (AI-CAD) for breast cancer detection on mammograms.
Materials And Methods:
Mammograms from 269 women with breast cancer were analyzed. Cancer visibility was determined based on reports from experienced radiologists. Two expert radiologists assessed mammographic findings and breast imaging reporting and data system (BI-RADS) categories by consensus for cases of visible cancer. AI-CAD results were reviewed to determine whether AI-CAD correctly marked the cancer site. AI-CAD detection rates were analyzed according to mammographic findings, BI-RADS categories, lesion size, histologic grade, lymph node involvement, and stage. The concordance between the findings of AI-CAD and those of experienced radiologists was also assessed. Mammographically occult cases were defined as those with negative mammographic findings by two radiologists.
Results:
AI-CAD detected 81.4% (219/269) of cancers, with higher detection rates occurring for larger lesion sizes, high histologic grades, lymph node involvement, and advanced stages. AI-CAD detection rates were higher for architectural distortion, mass, and calcification, but lower for asymmetry. Detection rates increased with higher BI-RADS categories and a higher number of mammography findings. Concordance between the assessment of AI-CAD and that of experienced radiologists was 88.5% (238/269). AI-CAD correctly detected 19.4% (6/31) of mammographically occult cases.
Conclusion:
AI-CAD detected 81.4% of cancers, with substantial concordance with the findings of experienced radiologists. It correctly identified 19.4% of mammographically occult cases.
