Modeling false positive error making patterns in radiology trainees for improved mammography education.

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Medical Education Technology

Background:

• Mammography is crucial for early breast cancer detection but suffers from high false positive rates.
• False positives in mammography lead to unnecessary patient anxiety and healthcare costs.
• Improving radiology trainee performance is essential to mitigate these diagnostic errors.

Purpose of the Study:

• To develop and evaluate a computational algorithm for modeling false positive error-making in radiology trainees.
• To identify specific locations in mammograms likely to cause false positive errors for trainees.
• To enhance radiology education by focusing on areas of common trainee difficulty.

Main Methods:

• A three-step algorithm was developed: Difference of Gaussian filtering and segmentation to identify suspicious regions, extraction of 133 features per region, and a random forest classifier trained on trainee data.
• The algorithm predicts the likelihood of a trainee making a false positive error based on their past annotations.
• The model was validated using a reader study involving 10 trainees and 3 experts interpreting 100 mammographic cases.

Main Results:

• The algorithm accurately identified locations prone to trainee false positive errors, outperforming random selection.
• With one predicted location per trainee, accuracy was 40% (vs. 0% random).
• With ten predicted locations per trainee, accuracy was 12% (vs. 0% random).

Conclusions:

• This study demonstrates the feasibility of using computer models to predict trainee false positive errors in mammography.
• The developed algorithm can identify specific error-prone locations in unseen mammograms.
• Targeted educational interventions based on these predictions may improve trainee performance and reduce false positives.