French researchers have developed voice biomarkers to estimate the level of excessive daytime sleepiness of a patient. Integrated into a virtual medical assistant, the new tool aims to facilitate home monitoring of patients.
Excessive daytime sleepiness is an unwanted and sometimes uncontrollable need to sleep during the day and corresponds to a momentary decrease in wakefulness. This is why, on a daily basis, it represents an inconvenience or even a risk of road, work or domestic accidents. To detect and measure it, several methods are used, in particular to have patients fill out questionnaires or to perform medical tests to diagnose a potential sleep pathology. But these methods have their limits, since in the first case, the answers are based on a subjective approach and in the second case, the tests can be long and expensive.
Researchers from the CNRS* are currently working on a tool that could be considered as complementary. They propose a much simpler and faster method of objective measurement of excessive daytime sleepiness, based on the analysis of the patient's voice. The researchers worked from a database of voice recordings of more than 120 patients, supplemented by the results of sleepiness measurements carried out by the usual methods, i.e. questionnaires and the recording of the patient's brain activity by electroencephalography, which is considered to be the reference test for objectifying the daytime sleepiness of a person.
Some biomarkers detected in the voice of sleepy people
A first step was to identify relevant markers of excessive daytime sleepiness in these recordings. "We had to identify markers that are both sensitive to drowsiness and specific: they are not influenced by other factors, such as age, gender, or patient anxiety. They are not influenced by other factors, such as age, gender, or patient anxiety," says Vincent Martin, a CNRS researcher at the Laboratoire Bordelais de Recherche en Informatique. The biomarkers (measurable characteristics that indicate a normal or pathological biological process) selected reflect the acoustic quality of the voice (frequency, voice intensity, etc.) or are related to the patient's reading quality (reading errors, position and length of pauses, etc.).
To develop their tool for measuring excessive daytime sleepiness, the researchers used machine learning techniques, testing their algorithms on real cases of patients already recorded in the database. "The resulting system is now able to identify different types or levels of sleepiness. The first results on three types of sleepiness will be published soon. ", they note. The researchers will now work on a database of healthy speakers to further validate the biomarkers. At the same time, they plan to turn this tool into software that can be used by other research teams.
Ultimately, the researchers hope that this new methodology for measuring excessive daytime sleepiness could be integrated into a virtual medical assistant, with the aim of monitoring the progress of patients at home in their natural context. It should be noted that, as the Institut National du Sommeil et de la Vigilance (INSV) reminds us, in Europe, drowsiness at the wheel is one of the leading causes of fatal traffic accidents. Traffic accidents caused by a driver falling asleep at the wheel are particularly severe and often fatal, due to the uncontrolled speed of the vehicle at the time of impact, and the inability of the driver to brake.
*CNRS/University of Bordeaux, the Bordeaux Computer Research Laboratory and the Sleep, Addiction and Neuropsychiatry Laboratory.
0 Comments