A novel speech analysis algorithm to detect cognitive impairment in a Spanish population

Area of Science:

• Neurology
• Computational Linguistics
• Gerontology

Background:

• Early detection of cognitive impairment is vital for timely diagnosis and care in the elderly.
• There is a need for brief, cost-effective screening tools to identify individuals requiring further cognitive evaluation.
• This study explores a novel screening technology utilizing automated voice analysis in a Spanish population.

Purpose of the Study:

• To evaluate the preliminary efficacy of an automated speech analysis algorithm for detecting cognitive impairment in Spanish speakers.
• To assess the performance of machine learning models trained on speech and task features for identifying cognitive decline.
• To establish the potential of voice analysis as a cost-effective screening method for cognitive impairment.

Main Methods:

• 174 Spanish-speaking individuals (cognitively normal, mild cognitive impairment, dementia) participated.
• Participants completed four language tasks, with recordings analyzed using text-transcription and digital-signal processing.
• Machine learning algorithms were developed and validated to detect cognitive impairment based on speech and task features.

Main Results:

• The automated speech analysis algorithm demonstrated significant differences in scores across cognitive groups (p < 0.01).
• The final algorithm achieved high accuracy (88.4%), sensitivity (87.5%), and specificity (89.2%) in distinguishing cognitively normal from impaired individuals.
• Area Under the Curve (AUC) values of 0.93 and 0.90 were obtained when comparing cognitively normal with impaired (MCI + dementia) and MCI groups, respectively.

Conclusions:

• The automated speech analysis algorithm shows initial promise as a screening tool for cognitive impairment in Spanish-speaking populations.
• Further validation in larger, diverse clinical populations is necessary to confirm the technology's utility.
• This approach offers a potential low-cost, accessible method for early cognitive decline detection.