Medical diagnostic systems: a case for neural networks

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Neurology

Background:

• Computer technology advances provide tools for medical data management and analysis.
• Artificial intelligence (AI), including rule-based and knowledge-based systems, is being explored for intelligent medical diagnostics.
• Artificial neural networks (ANN) offer a promising approach to enhance diagnostic capabilities without replacing physician decision-making.

Purpose of the Study:

• To introduce artificial neural networks (ANN) as a tool for developing an intelligent diagnostic system.
• To demonstrate a methodology for an integrated diagnostic system for neuromuscular disorders.
• To enhance physician capabilities in reaching accurate diagnoses.

Main Methods:

• Developed an integrated diagnostic system using modules for clinical examination and laboratory test data.
• Standardized examination procedures with expert protocols to create numerical data vectors.
• Utilized unsupervised self-organizing feature maps algorithm to develop ANN models trained on data from 41 subjects and tested on 30 subjects.

Main Results:

• ANN models trained with clinical data achieved diagnostic yields of 73-93% for unknown cases.
• Models trained with combined clinical and laboratory data showed diagnostic yields of 73-100%.
• Self-organized feature maps provided a user-friendly interface for physicians, aiding in disease progression monitoring.

Conclusions:

• Artificial neural networks (ANN) are effective tools for building intelligent diagnostic systems for neuromuscular disorders.
• Integrating clinical and laboratory data significantly improves diagnostic accuracy.
• The developed system offers a valuable human-computer interface to support clinical decision-making and patient monitoring.