Questionnaires for the Assessment of Cognitive Function Secondary to Intake Interviews in In-Hospital Work and...

Area of Science:

• Gerontology
• Computational Linguistics
• Artificial Intelligence

Background:

• Dementia prevalence is rising globally, necessitating efficient early detection methods.
• Current screening tools for dementia and mild cognitive impairment (MCI) are often costly and time-consuming.
• Speech pattern analysis offers a potential avenue for accessible and inexpensive cognitive screening.

Purpose of the Study:

• To develop and evaluate a machine learning model for classifying older adults with dementia and MCI using speech patterns.
• To assess the effectiveness of a standardized 30-question intake questionnaire in cognitive assessment.
• To compare the performance of Mel-spectrogram and Mel-Frequency Cepstral Coefficients (MFCC) in speech-based cognitive classification.

Main Methods:

• A cohort of 29 older adults (aged 72-91) was recruited, with cognitive status assessed using the Mini-Mental State Examination (MMSE).
• Participants were categorized into moderate dementia (MMSE ≤ 20), mild dementia (MMSE 21-23), and MCI (MMSE 24-27).
• Machine learning models were trained using acoustic features extracted from speech samples, comparing Mel-spectrogram and MFCC.

Main Results:

• Mel-spectrogram analysis demonstrated superior performance over MFCC across all classification tasks, yielding higher accuracy, precision, recall, and F1-scores.
• The multi-classification model using Mel-spectrogram achieved a peak accuracy of 0.932.
• Binary classification of moderate dementia versus MCI using MFCC resulted in the lowest accuracy (0.502), though false positive rates were generally low.

Conclusions:

• Speech pattern analysis, particularly using Mel-spectrograms, shows significant potential for accurate and cost-effective early detection of dementia and MCI.
• The developed questionnaire and machine learning approach offer a feasible screening method for cognitive impairment in older adults.
• Further research is needed to address the relatively high false negative rates observed in certain classifications.