Case-based explanation for medical diagnostic programs, with an example from gynaecology

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Clinical Decision Support Systems

Background:

• Machine assistance in medical diagnosis often relies on retrieving similar past cases.
• Nearest neighbors classification can be less accurate than other statistical methods.
• There is a need to improve the accuracy of case-based reasoning in diagnostics.

Purpose of the Study:

• To combine the transparency of nearest neighbors with the accuracy of statistical methods.
• To develop a novel similarity metric for medical case comparison.
• To evaluate the method's effectiveness in diagnosing gynecological abdominal pain.

Main Methods:

• Utilized Bayes' theorem to define a new metric for case similarity, assuming conditional independence.
• Employed a case study focusing on abdominal pain of suspected gynecological origin.
• Compared the performance of the new metric against traditional methods like Hamming distance.

Main Results:

• The Bayes' theorem-based metric demonstrated strong correspondence with clinical notions of similarity.
• This new metric significantly increased classification accuracy compared to standard nearest neighbors.
• Achieved diagnostic accuracy comparable to the Bayes' method itself.

Conclusions:

• Integrating statistical methods like Bayes' theorem enhances nearest neighbors' diagnostic accuracy.
• The developed similarity metric offers a transparent yet accurate approach for machine-assisted diagnosis.
• This hybrid method shows promise for improving clinical decision support systems.