Multimodal Dementia Screening from Brain Magnetic Resonance Imaging and Conversational Speech

Area of Science:

• Neuroimaging
• Computational Linguistics
• Biomarker Discovery

Background:

• Early detection of Alzheimer's dementia is critical for timely symptomatic treatment.
• Biomarkers from Magnetic Resonance Imaging (MRI) and spoken communication show promise for dementia screening.

Purpose of the Study:

• To compare the screening performance of unimodal (MRI or speech) versus multimodal (combined MRI and speech) biomarkers.
• To evaluate different data fusion techniques (Early, Late, Tensor Fusion) for multimodal analysis.
• To assess the impact of training set size on unimodal screening performance.

Main Methods:

• Utilized structural MRI data and interview recordings from the Interdisciplinary Longitudinal Study on Adult Development and Aging.
• Developed and compared unimodal and multimodal classifiers for state and predictive screening of cognitive impairment.
• Assessed classifier performance using Unweighted Average Recall, comparing Early, Late, and Tensor Fusion methods.

Main Results:

• Multimodal classifiers using Early Fusion significantly outperformed unimodal classifiers for both state (83.2% UAR) and predictive screening (74.9% UAR).
• Speech-derived markers demonstrated comparable utility to MRI-derived markers for predictive screening in this dataset.
• The study analyzed the effect of varying training set sizes on unimodal screening performance.

Conclusions:

• Combining MRI and speech biomarkers through early fusion enhances dementia screening accuracy.
• Speech analysis is a valuable, complementary tool for predictive screening of cognitive impairment, on par with MRI.
• Multimodal approaches represent a promising avenue for early and effective Alzheimer's dementia detection.