- Статьи
- Internet and technology
- Speech will find: voice signs of diseases will help AI to conduct a "medical examination" by phone
Speech will find: voice signs of diseases will help AI to conduct a "medical examination" by phone
Russian scientists have discovered speech signs of chronic heart failure, asthma, diabetes mellitus and other non-communicable diseases. These markers can be used for mass remote monitoring of public health, for example, using automatic phone calls, the results of which are processed by AI. The collected biometric data is planned to be stored in a biobank, which will be accessed by medical centers throughout Russia. Experts note that testing on tens of thousands of real patients will be required to assess the accuracy of the method.
Diagnosis of diseases by phone
Specialists from Samara State Medical University (SamSMU) have proposed an AI-based technology for mass screening of chronic diseases over the phone. For example, it can be a robotic call to potential patients. Since the implementation of the method will require processing a large amount of data and significant computing power, doctors suggest using the capabilities of the blockchain when the load is distributed among various network participants. This approach will eliminate the use of expensive equipment and make the approach economically feasible.
After analyzing 4,500 anonymized audio recordings, scientists found speech parameters indicating chronic heart failure. Speech biomarkers for bronchial asthma, COPD, hypertension, diabetes mellitus and other pathologies were also identified.
— Our published work is the world's first study of speech signal parameters in Russian—speaking patients with chronic non-communicable diseases. We consider it promising to conduct research on multicenter validation (verification of data accuracy by several centers. — Izvestia) clinically significant biomarkers for the most common and socially significant non-communicable pathologies. In the future, they could be included in national recommendations on screening, diagnosis and dynamic remote follow—up of patients," said Andrey Garanin, director of the Scientific and Practical Center for Remote Medicine at SamSMU, associate professor of Outpatient Care with a course in telemedicine.
The researchers noted that the dynamics of speech biomarkers in patients correlated with the progression of each pathology studied. Neural network training based on the dataset of speech features collected by scientists also yielded high results.
For example, in chronic heart failure, the acoustic parameters and phonetic characteristics of speech change. This is due to fluid retention in the tissues, which increases the viscosity of the vocal folds, as well as a weakening of the vagus nerve and a change in the dynamics of airflow. These physiological mechanisms accompany the syndrome and directly affect the process of speech formation.
Data on quantitative changes in a number of vocal biomarkers can become the basis for remote monitoring of patients' condition after discharge from the hospital. This approach makes it possible to obtain objective indicators along with the patient's subjective feelings, as well as to send them for an in-depth examination in a timely manner. At the same time, monitoring does not require specialized equipment or the participation of qualified personnel.
Will AI help to make a diagnosis
— We propose to organize a network structure for remote monitoring of patients or healthy people using the blockchain type, so that all medical centers have access to the national biobank. It would contain the results of biometrics. For example, a pulse wave can be measured using a video image of a face. You can also analyze the voice, the iris, and the skin of the forehead," said Andrey Garanin.
To implement this approach, powerful data centers are really needed. Audio, video, and photographic materials of patients create huge amounts of information. Theoretically, they can be processed using quantum computers, but they are too expensive. Therefore, it is better to use intranet computing for these operations, when many classic computers are combined into one system, as is done, for example, during mining, the specialist added.
An experienced doctor may notice signs of abnormalities in the voice without additional examination. However, in order to adequately assess the accuracy of AI diagnostics over the phone, it will be necessary to test the system on tens of thousands of patients, explained Andrey Prodeus, professor, chief freelance pediatric allergist and immunologist at the Ministry of Health of the Moscow Region.
— There are characteristic signs of shortness of breath in people with cardiovascular insufficiency and bronchial asthma. However, in practice, no one can say yet how accurately AI will be able to determine this. There is no particular harm if the system once again sends someone for a consultation, but what if the patient's voice does not betray the pathology, but he has it? Who will be responsible for this? — the doctor noted in a comment to Izvestia.
According to Maxim Kolyasnikov, associate professor at the UrFU Institute of Economics and Management and the Department of Future Technologies at MIPT, the presented solution is more of a promising platform than a ready—made mass screening tool.
— From an economic point of view, such developments are important primarily as an inexpensive tool for pre-selection and remote monitoring of patients with chronic heart failure. With successful clinical validation, they can reduce the number of late referrals, repeated hospitalizations and the burden on the full—time healthcare sector," the expert believes.
According to him, such solutions may be of interest to large ecosystem companies that are already developing consumer health products. However, this is only possible if regulation, medical ethics, and patient consent are strictly followed.
Переведено сервисом «Яндекс Переводчик»
