Clinical Manifestations.

Area of Science:

• Neurology
• Computational Linguistics
• Digital Biomarkers

Background:

• Automated speech and language analysis (ASLA) is emerging as a digital biomarker for neurological conditions.
• Validation of ASLA metrics against established pathological and neurocognitive measures is limited.

Purpose of the Study:

• To correlate and compare ASLA metrics with pathological and neurocognitive measures in individuals with Alzheimer's dementia (AD), mild cognitive impairment (MCI), and nonfluent/agrammatic variant primary progressive aphasia (nfvPPA).

Main Methods:

• Assessed 226 participants (60 AD, 52 MCI, 24 nfvPPA, 100 healthy controls).
• Extracted speech timing and word property metrics from verbal fluency and paragraph reading tasks.
• Employed inferential statistics and machine learning to assess metric sensitivity, correlation with brain changes, diagnostic comparability, and predictive power for pathology.

Main Results:

• ASLA metrics differentiated individuals with AD from controls (AUC=.89), correlating with temporo-frontal atrophy and default-mode network hypoconnectivity.
• Speech timing features showed similar classification performance (AUC=.83) and were non-inferior to standard cognitive (AUC=.81) and neuroimaging (AUC=.89) measures.
• ASLA successfully classified MCI (AUC=.80) and nfvPPA (AUC=.95), predicting brain changes and, in nfvPPA, autopsy-confirmed pathology years in advance.

Conclusions:

• ASLA metrics serve as sensitive early markers for AD and related conditions, predicting brain anomalies and pathology.
• ASLA performance is comparable to standard neuropsychological and neuroimaging assessments, meeting non-inferiority criteria for digital tests.
• The affordability and scalability of ASLA support its role in promoting equitable dementia screening and assessment.