[Formation of diagnostic rules by using neural networks].

Area of Science:

• Artificial Intelligence in Medicine
• Computational Neuroscience
• Clinical Decision Support Systems

Background:

• Traditional diagnostic methods can be complex and time-consuming.
• Developing accurate and interpretable AI models for healthcare remains a challenge.
• Bridging the gap between complex AI algorithms and clinical usability is crucial.

Purpose of the Study:

• To develop and validate neural networks for medical classification tasks.
• To create interpretable AI models that mimic clinical diagnostic reasoning.
• To enhance the reliability and validity of AI-assisted medical diagnoses.

Main Methods:

• Training neural networks on clinician-proposed, unrepresentative patient datasets.
• Minimizing the number of signs and neurons for model efficiency.
• Encoding estimates using two intervals for validity control.
• Representing neural network decisions as logical formulas and diagnostic tables.

Main Results:

• Neural networks correctly classified all presented cases.
• The developed logical formulas were easily interpretable and tabular.
• AI-driven decisions showed over 88% agreement with expert medical conclusions.
• The system provided affinity estimates to control the validity of diagnostic outputs.

Conclusions:

• Minimized neural networks can generate interpretable logical formulas for medical diagnosis.
• The proposed AI approach effectively supports clinical decision-making with high accuracy.
• This method offers a reliable and valid tool for physicians, enhancing diagnostic processes.