Cognitive decline assessment using semantic linguistic content and transformer deep learning architecture.

Area of Science:

• Computational linguistics and artificial intelligence applied to healthcare.
• Neuroscience and cognitive science research into neurodegenerative diseases.

Background:

• Dementia is a progressive cognitive decline impacting daily activities and caretaking needs.
• Early dementia detection is crucial for reducing caregiving burden and improving patient outcomes.
• Speech and language difficulties are recognized as early indicators of dementia.

Purpose of the Study:

• To develop and assess an approach for cognitive decline assessment using speech data.
• To specifically evaluate speech relevance as an indicator of memory recall in dementia.
• To leverage deep learning models for enhanced dementia detection through linguistic analysis.

Main Methods:

• A cross-sectional, online, self-administered study utilizing deep learning transformers (BERT, Sentence-Transformer).
• Speech transcripts were encoded to derive contextually descriptive representations.
• Cosine similarity metrics were employed to measure the semantic relevance of uttered sentences, compared against the Pitt Corpus Dementia dataset.

Main Results:

• The proposed multi-QA-MPNet model demonstrated superior performance in capturing context and semantic memory.
• Cosine similarity was identified as the most effective measure for evaluating the relevance of spoken language sequences.
• The approach showed significant potential in analyzing speech for early dementia indicators.

Conclusions:

• Cosine similarity metrics can effectively identify semantic relevance in spoken language, serving as an early warning sign for dementia.
• Speech analysis, alongside other modalities like brain imaging, can aid in early dementia recognition.
• The developed deep learning approach offers significant clinical implications for improving dementia diagnosis accuracy and efficiency.