Biomarkers.

Area of Science:

• Neuroscience
• Biomarkers
• Speech Analysis

Background:

• Speech features from Logical Memory Delay Recall (LMd) tests are linked to early tau burden.
• Existing software tools for speech feature extraction use varied methodologies.
• Validating the robustness of these speech features is crucial for reliable biomarker development.

Purpose of the Study:

• To compare speech features generated by two distinct software tools (ki:elements and CLAN).
• To assess the association between these speech features and tau pathology.
• To validate the robustness and consistency of automated speech feature extraction methods.

Main Methods:

• Analysis of data from 237 Framingham Heart Study participants with PET imaging and LMd tests.
• Extraction of five speech features using ki:elements' SIGMA platform and CLAN software.
• Correlation analysis (Spearman) and multiple linear regression to examine associations with tau SUVR, adjusting for covariates.

Main Results:

• All speech features extracted by both tools showed significant correlations (p < 0.001).
• Consistent patterns were observed in the association between speech features and tau SUVR across brain regions.
• Longer between-utterance pause duration correlated with higher tau SUVR in entorhinal, inferior temporal, and inferior parietal regions.

Conclusions:

• Speech features, including automated ones, demonstrate compatible associations with early tau burden across different extraction methods.
• Automated speech analysis holds promise as a scalable biomarker for neurodegenerative diseases.
• Further investigation into automated linguistic and semantic features is warranted for enhanced diagnostic capabilities.