Using Contextual Learning to Improve Diagnostic Accuracy: Application in Breast Cancer Screening.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Oncology

Background:

• Clinical decision-making for patients at risk of breast cancer requires accurate diagnostic tools.
• Current breast cancer screening methods face challenges with false positives and negatives.

Purpose of the Study:

• To present a novel clinical decision support tool for improving breast cancer diagnosis.
• To enhance the specificity of breast cancer screening and minimize diagnostic errors.

Main Methods:

• A clinical decision support system utilizing an online contextual learning algorithm.
• Integration of patient clinical context (demographics, medical history) to refine diagnostic strategies.
• Analytical evaluation of diagnostic performance and feature relevancy assessment.

Main Results:

• The proposed algorithm's false positive rate asymptotically converges to the optimum.
• Individualized confidence estimates for diagnosis recommendations are provided.
• Outperformed current clinical practice by 36% in reducing false positive rates at a 2% false negative rate.

Conclusions:

• The novel decision support tool significantly improves breast cancer diagnostic accuracy.
• The system effectively minimizes false positives while avoiding false negatives.
• Contextual feature analysis provides valuable insights for reducing diagnostic errors.