[Expert systems and automatic diagnostic systems in histopathology--a review]

Area of Science:

• Digital pathology and medical informatics
• Artificial intelligence in healthcare
• Computational pathology

Background:

• Pathological information systems are increasingly digitized in Japanese hospitals, with increasing use of tele-pathology.
• Diagnostic pathology often relies on individual pathologists, highlighting the need for decision support.
• Existing knowledge-based expert systems are primarily educational, not diagnostic aids.

Purpose of the Study:

• To review knowledge-based expert systems for diagnostic support.
• To present a 3-year development experience of an artificial neural network-based pathological image classifier.
• To explore computer-assisted tools for improving pathological diagnosis quality.

Main Methods:

• Review of literature on knowledge-based interactive expert systems.
• Development of an image-analysis-based automatic histopathological image classifier using artificial neural networks.
• Evaluation of the classifier's performance on breast tumor patterns.

Main Results:

• The developed classifier provides a 'fitting rate' for specific diagnostic patterns, such as 'fibroadenoma pattern'.
• Artificial neural networks technology was utilized for pathological image classification.
• The system aims to assist pathologists, particularly solo practitioners.

Conclusions:

• Computer-assisted diagnosis systems can serve as valuable tools for quality assurance in pathology.
• Image-analysis-based classifiers show promise for automatic histopathological image analysis.
• Further development is needed to support solo pathologists in diagnostic decision-making.