Intelligent dialogue based on statistical models of clinical decision-making

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Decision Support Systems
• Bayesian Networks

Background:

• The independence Bayesian model is frequently utilized in clinical decision support software.
• Effective reasoning strategies are needed for these programs to engage in clinically relevant dialogue and articulate their diagnostic processes.

Purpose of the Study:

• To develop and implement a reasoning strategy for Bayesian diagnostic programs.
• To enhance clinical decision support by enabling programs to conduct pertinent dialogue and explain their reasoning.
• To apply this strategy to a program for diagnosing acute abdominal pain.

Main Methods:

• Developed a reasoning strategy for Bayesian models.
• Implemented the strategy in a diagnostic program for acute abdominal pain, utilizing the de Dombal et al. Bayesian model.
• Incorporated artificial intelligence techniques such as shared initiative and critiquing for dialogue design.
• Employed a flexible, goal-driven strategy to confirm diagnoses or rule out alternatives.
• Selected symptoms and signs based on their expected weights of evidence.

Main Results:

• The implemented program effectively diagnoses acute abdominal pain.
• The reasoning strategy enables clinically pertinent dialogue and clear explanations.
• The system successfully confirms clinician diagnoses or rules out likely alternatives.
• Symptom and sign selection is optimized by expected weights of evidence.

Conclusions:

• The developed reasoning strategy significantly improves the functionality of Bayesian diagnostic programs.
• This approach enhances clinical decision support by facilitating interactive and explanatory diagnostic processes.
• The program provides a valuable tool for diagnosing acute abdominal pain.