JMIR Publications

ABSTRACT

Background:

The rise in life expectancy is associated with an increase of long-term and gradual cognitive decline. Since treatment effectiveness enhances at early stage of the disease, there is the need to find low-cost and ecological solution for mass-screening of community-dwelling elderly people.

Objective:

This work aimed at exploiting automatic analysis of free speech to identify signs of cognitive function decline.

Methods:

A sample of 266 subjects aged over 65 years in Italy and Spain were recruited and divided into three groups according to their Mini Mental Status Examination (MMSE) score. People were asked to tell a story and to describe a picture and voice recordings were used to automatically extract high-level features on different time scales. Based on these features, machine learning algorithms were trained to solve binary and multi-class classification problems, using both mono- and cross-lingual approaches. The algorithms were enriched by the use of SHAP for model explainability.

Results:

On the Italian dataset, healthy subjects (MMSE≥27) were automatically discriminated from subjects with a mildly impaired cognitive function (20≤ MMSE≤26) and from those with a moderate to severe impairment of cognitive function (11≤MMSE≤19) with an accuracy of 80% and 86%, respectively. Slightly lower performances were achieved on the Spanish and multi-language dataset.

Conclusions:

This work proposed a transparent and unobtrusive assessment method, which might be included in a mobile app for large-scale monitoring of the cognitive functionality in elderly people. Voice confirmed to be an important biomarker of cognitive decline due to its non-invasive and easily accessible nature. Clinical Trial: Not applicable