Mild Cognitive Impairment Detection System Based on Unstructured Spontaneous Speech: Longitudinal Dual-Modal...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Gerontology

Background:

• Cognitive diseases, including Alzheimer's, are increasing with global population aging.
• Mild cognitive impairment (MCI) is a critical transitional stage requiring early diagnosis to slow disease progression.
• Early detection of MCI is vital for timely treatment and managing healthcare costs.

Purpose of the Study:

• To develop a dual-modal longitudinal cognitive detection system for MCI using autobiographical memory (AM) speech data.
• To enhance the accuracy of MCI detection by analyzing both speech and text data.
• To track cognitive changes over time in spontaneous speech through an aging trajectory module.

Main Methods:

• Utilized autobiographical memory (AM) test speech data for a dual-modal analysis (speech and text).
• Introduced an aging trajectory module with local and global alignment loss functions to capture time-related cognitive changes.
• Developed a longitudinal detection system to monitor cognitive status over multiple time points.

Main Results:

• The longitudinal model with the aging trajectory module achieved AUCs of 0.85 and 0.89 on two Chinese datasets.
• Demonstrated significant improvement over cross-sectional, single time point models.
• Validated the model's generalizability on the ADReSSo dataset, achieving accuracy over 0.88.

Conclusions:

• Presented a noninvasive, scalable approach for early MCI detection using longitudinal AM speech data.
• The dual-modal system with an aging trajectory module effectively captures cognitive decline trends.
• The method shows robustness and generalizability for real-world, long-term cognitive monitoring.