Interpretable Recognition for Dementia Using Brain Images.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Machine Learning

Background:

• Machine learning models excel in neuroimage-based dementia recognition but often lack interpretability.
• Interpretability is crucial for assessing model confidence in clinical applications.
• Takagi-Sugeno-Kang (TSK) fuzzy classifiers offer high interpretability and good performance but can struggle with high-dimensional data complexity.

Purpose of the Study:

• To improve the interpretability of neuroimage-based dementia recognition models.
• To address the complexity issue in Takagi-Sugeno-Kang (TSK) fuzzy classifier antecedents when handling high-dimensional data.
• To develop a model that balances high recognition performance with concise, interpretable fuzzy rules.

Main Methods:

• Introduced subspace clustering technique for antecedent learning in Takagi-Sugeno-Kang (TSK) fuzzy classifiers.
• Applied the developed model to neuroimage-based dementia recognition tasks.
• Focused on generating concise fuzzy rule antecedents to maintain interpretability.

Main Results:

• The proposed model demonstrated promising recognition performance in dementia detection.
• The subspace clustering approach successfully generated concise fuzzy rule antecedents.
• The enhanced interpretability of the fuzzy rules aids in understanding the model's reasoning process.

Conclusions:

• The integration of subspace clustering with Takagi-Sugeno-Kang (TSK) fuzzy classifiers offers an effective approach for interpretable neuroimage-based dementia recognition.
• The method successfully addresses the challenge of complex antecedents in high-dimensional data, enhancing model transparency.
• This approach provides a valuable tool for building more trustworthy and understandable AI models in medical diagnostics.