Clinical validation of a voice-based AI tool for screening cognitive impairment: a prospective multicenter study

Area of Science:

• Neurology
• Gerontology
• Medical Technology

Background:

• Traditional cognitive screening methods are limited in scalability.
• There is a need for early and accessible screening tools for cognitive impairment.

Purpose of the Study:

• To assess the diagnostic accuracy and usability of SPICK, a Korean voice-based cognitive screening tool.
• To determine if SPICK meets predefined sensitivity and specificity thresholds for detecting cognitive impairment.

Main Methods:

• A prospective, multicenter validation study involved 399 participants.
• SPICK's results were compared against a reference diagnosis made by neurologists.
• Key metrics included sensitivity, specificity, accuracy, AUC, and System Usability Scale (SUS).

Main Results:

• SPICK achieved 85.71% sensitivity and 74.29% specificity for cognitive impairment detection, exceeding thresholds.
• Area Under the Curve (AUC) was 0.800, with an overall accuracy of 83.71%.
• Mean SUS score was 67.32, indicating acceptable usability.

Conclusions:

• SPICK demonstrates clinically significant accuracy and acceptable usability.
• The tool has potential as an automated, voice-based screening solution for various clinical environments.
• SPICK supports early detection of cognitive impairment, aiding timely intervention.