Improving Alzheimer's Disease Detection for Speech Based on Feature Purification Network

Area of Science:

• Neuroscience
• Computational Linguistics
• Artificial Intelligence

Background:

• Alzheimer's disease (AD) diagnosis relies on subjective and time-consuming clinical methods.
• Current diagnostic criteria for AD have limitations in speed, cost, and objectivity.
• Language analysis shows promise for objective and efficient AD detection.

Purpose of the Study:

• To develop an advanced feature purification network to enhance transformer models for AD diagnosis.
• To improve the discriminative power of features extracted by transformer models.
• To achieve state-of-the-art classification results on dementia datasets.

Main Methods:

• Proposed a novel feature purification network integrated with transformer models.
• Filtered out common, non-indicative features from traditional transformer outputs.
• Applied the enhanced model to three public dementia datasets.

Main Results:

• Achieved markedly improved classification results on dementia datasets.
• Demonstrated superior performance of the feature purification network in enhancing transformer capabilities.
• Attained state-of-the-art (SOTA) results on the Pitt dataset for AD classification.

Conclusions:

• The proposed feature purification network significantly enhances transformer-based language analysis for AD diagnosis.
• This approach offers a more accurate and potentially efficient method for identifying Alzheimer's disease.
• The findings suggest a promising direction for leveraging AI in neurodegenerative disease diagnostics.