- Статьи
- Science and technology
- Voice communication: AI platform will detect diabetes and heart disease by speech
Voice communication: AI platform will detect diabetes and heart disease by speech
Russian scientists have created an AI platform capable of assessing the risk of cardiovascular and metabolic diseases by voice. The prototype of the application analyzes a short audio recording made on a smartphone and outputs the result in a few seconds. According to the authors of the project, the technology can be used for mass screening, remote monitoring of patients and epidemiological studies. The experts interviewed by Izvestia noted the usefulness of the development, but stressed that the voice can only act as one of the potentially useful markers — the program will not replace a doctor's diagnosis.
A platform for detecting diseases by voice
A team of graduates from the Sechenov University Digital Department has developed a prototype of a unique AI-based application that allows assessing the risk of cardiovascular and metabolic diseases using voice biomarkers.
The application automatically offers the user instructions for recording an audio file, analyzes the voice signal within a few seconds and outputs the result. For testing, it is enough for the user to record a short audio file using a smartphone.
As Denis Kuznetsov, the author of the project, a graduate of the Digital Department, Candidate of Biological Sciences, explained to Izvestia, the respiratory system, the muscles of the larynx and articulatory apparatus, as well as the mechanisms of nervous regulation, are involved in the process of speech.
With changes in the body's condition, for example, with inflammatory processes, metabolic disorders, or increased stress on the cardiovascular system, voice parameters may change. These changes are almost indistinguishable by ear, but they can be detected using digital audio processing and machine learning methods.
— We have created a prototype platform that analyzes voice patterns in order to identify a predisposition to the development of cardiometabolic diseases — arterial hypertension, coronary heart disease, type 2 diabetes mellitus and others, — said the developer. — In the future, this technology can be used for preliminary risk assessment during medical examinations, remote monitoring of patients' condition, as well as for predicting the burden on the healthcare system.
Unlike its foreign counterparts, which are still at the registration stage, this solution is adapted to the peculiarities of Russian-language speech and supplemented with unique features that may later be useful for epidemiological research, the authors of the project indicated.
How the SIRIUS project helped in the development of the platform
The application's algorithms are trained on multicenter clinical patient data, as well as on audio recordings of healthy volunteers collected during the annual SIRIUS2023 experiment. The total volume of the database exceeds 4 thousand unique audio files. Sensitivity and specificity for most models exceed 81% and 90%.
The team's future plans include developing a machine learning model for predicting the condition of patients, collecting data from cohorts (large groups) of patients and pilot studies in several clinics, and registration events.
The development of a platform for assessing cardiometabolic risks by voice is a promising area at the intersection of medicine and digital technologies, says Philip Kopylov, the project's scientific director, Director of the Institute of Personalized Cardiology at Sechenov University.
The physiological relationship between voice parameters and the state of the cardiovascular system is scientifically proven, and the non-invasiveness of the method opens up opportunities for mass screening and remote monitoring. At the same time, the claimed high accuracy rates require independent validation (quality assessment conducted by a disinterested third party) in real clinical settings.
The voice really reflects the work of the respiratory system, autonomous nervous regulation, muscle tone, and indirectly metabolic and inflammatory status, Albert Rizvanov, head of the Personalized Medicine Excellence Center at the Institute of Fundamental Medicine and Biology at Kazan Federal University, told Izvestia. This has been well demonstrated in a number of experimental and clinical studies.
—However, cardiometabolic diseases are by definition multifactorial — they depend on genetics, age, gender, lifestyle, medications, and concomitant diseases," the specialist noted. — Therefore, voice can act as only one of the weak but potentially useful predictors (predictors of symptoms. In the same way as measuring body temperature with a thermometer helps to suspect an infection, but not to diagnose it.
The claimed sensitivity and specificity indicators (81-90%) look plausible for pilot models, but without independent external validation in other cohorts and in real clinical scenarios, the actual accuracy is inevitably overestimated, Albert Rizvanov concluded. The value of the platform lies in pre—sorting risks, epidemiological studies and monitoring of large populations, where simplicity, cheapness and non-invasiveness are important.
The voice is a complete physiological signal, which is formed by respiration, the autonomic nervous system, muscle tone and metabolism in the body. All these mechanisms are closely related to the molecular processes: inflammation, oxidative stress, hormonal regulation and energy metabolism underlying cardiovascular diseases, confirmed by molecular biologist Arina Kholkina.
— Changes at the level of cells and signaling pathways over time are reflected in the work of organs, and therefore in the parameters of the voice. The value of such platforms lies in their non—invasiveness and scalability — they can detect risk long before clinical manifestations," the specialist said. But the key stage will be the biological interpretation of the detected vocal patterns and their confirmation in independent clinical and molecular studies.
The development received a grant from the Foundation for the Promotion of Innovation, and also entered the final of the Open Innovation Window competition.
Переведено сервисом «Яндекс Переводчик»
