A diagnostic method that uses causal knowledge and linear programming in the application of Bayes' formula

Area of Science:

• Computer Science
• Medical Informatics
• Artificial Intelligence

Background:

• Bayes' formula is widely used in computer-based medical diagnostic systems.
• A common assumption of conditional independence among findings is often violated, leading to inaccurate diagnostic probabilities.

Purpose of the Study:

• To present a method for structuring medical findings based on causal knowledge and probabilistic dependencies.
• To develop an inference procedure for calculating accurate posterior probabilities of diagnostic hypotheses.

Main Methods:

• Utilizing causal knowledge to build a network of causally related findings.
• Implementing an inference procedure to propagate probabilities within this network.
• Employing a linear programming technique to bound propagated probabilities under known constraints.

Main Results:

• The proposed method allows for more accurate posterior probability assignments in diagnostic systems.
• Structuring findings by probabilistic dependencies mitigates errors from the conditional independence assumption.
• The linear programming technique provides bounds for propagated probabilities.

Conclusions:

• Causal knowledge can effectively structure findings in medical diagnostic systems.
• The developed inference procedure enhances the accuracy of diagnostic probability calculations.
• This approach offers a robust method for handling probabilistic dependencies in medical AI.