Automatic classification of AD pathology in FTD phenotypes using natural speech.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Speech Pathology

Background:

• Screening for Alzheimer's disease neuropathologic change (ADNC) in individuals with atypical presentations like frontotemporal dementia (FTD) is clinically challenging.
• Distinguishing ADNC from other pathologies, such as frontotemporal lobar degeneration (FTLD), is crucial for effective patient management.

Purpose of the Study:

• To develop and validate automatic speech-based classifiers for distinguishing FTD patients with ADNC from those with FTLD.
• To identify speech features and their corresponding neuroanatomical correlates associated with different pathologies.

Main Methods:

• Trained automatic classifiers using 99 speech features from 1-minute speech samples of 179 participants (36 ADNC, 60 FTLD, 89 healthy controls).
• Assigned neuropathology based on autopsy or cerebrospinal fluid (CSF) biomarkers.
• Utilized structural network-based magnetic resonance imaging (MRI) to analyze anatomical correlates of speech features.

Main Results:

• The classifier achieved an area under the curve (AUC) of 0.88 ± 0.03 for distinguishing ADNC from FTLD.
• The classifier achieved an AUC of 0.93 ± 0.04 for distinguishing patients from healthy controls.
• Noun frequency and pause rate correlated with gray matter volume loss in the limbic and salience networks, respectively.

Conclusions:

• Brief, naturalistic speech samples are effective for in vivo screening of FTD patients for underlying ADNC.
• This research supports the development of digital assessment tools for FTD, improving diagnostic capabilities.
• The findings highlight the potential of speech analysis combined with AI for non-invasive neuropathology screening.