Detecting Dementia Through Interactive Computer Avatars.

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Dementia detection often relies on traditional cognitive tasks and language analysis.
• Existing methods may not fully capture subtle early indicators of cognitive decline.
• There is a need for innovative, accessible tools for early dementia screening.

Purpose of the Study:

• To develop and evaluate a computer avatar system for automated dementia detection.
• To explore the efficacy of audiovisual features extracted from spoken dialogues for dementia prediction.
• To identify key speech and non-verbal cues indicative of dementia.

Main Methods:

• Developed a computer avatar system generating neuropsychological queries via spoken dialogue.
• Recorded spoken dialogues from 29 participants (14 dementia, 15 healthy controls).
• Extracted audiovisual features and applied machine learning algorithms (SVM, logistic regression) for classification.

Main Results:

• Support Vector Machines (SVM) outperformed logistic regression in dementia classification.
• The system achieved a high detection performance of 0.93 area under the ROC curve.
• Identified significant features including speech gaps, fundamental frequency variation, voice quality, and smiling ratio.

Conclusions:

• The developed spoken dialogue system shows significant potential for automated dementia detection.
• Audiovisual feature analysis from interactive dialogues can effectively identify individuals with dementia.
• This technology can serve as a valuable assistive tool for healthcare professionals in early dementia screening.