Using the ID3 algorithm to find discrepant diagnoses from laboratory databases of thyroid patients.

Area of Science:

• Artificial Intelligence in Medicine
• Machine Learning Applications
• Clinical Decision Support Systems

Background:

• Expert systems face challenges with rare cases in machine learning.
• Developing decision support systems (DSS) to encompass all clinical variations is difficult.
• Inductive learning programs can identify atypical cases.

Purpose of the Study:

• To construct an expert system using machine learning for detecting rare clinical cases.
• To evaluate the effectiveness of ID3 and pessimistic pruning algorithms in building a DSS.
• To create a feedback mechanism for clinicians based on system-identified misclassifications.

Main Methods:

• A decision support system (DSS) was developed using the ID3 and pessimistic pruning algorithms.
• The DSS was trained on patient record data to generate a decision tree.
• Misclassified cases were identified by comparing the decision tree's output with clinician diagnoses.

Main Results:

• The system achieved high accuracy in identifying misclassified cases (over 90%) with 95% specificity.
• Approximately 5-10% of functional thyroid disorder cases were found to be misclassified.
• The generated checklist provided valuable feedback to clinicians regarding potential errors.

Conclusions:

• Dynamic self-learning systems can enhance clinical decision-making quality control.
• Machine learning, specifically decision trees, can effectively assist in identifying challenging medical cases.
• This approach offers a practical method for improving the reliability of expert systems in healthcare.