-
Categories
-
Pharmaceutical Intermediates
-
Active Pharmaceutical Ingredients
-
Food Additives
- Industrial Coatings
- Agrochemicals
- Dyes and Pigments
- Surfactant
- Flavors and Fragrances
- Chemical Reagents
- Catalyst and Auxiliary
- Natural Products
- Inorganic Chemistry
-
Organic Chemistry
-
Biochemical Engineering
- Analytical Chemistry
- Cosmetic Ingredient
-
Pharmaceutical Intermediates
Promotion
ECHEMI Mall
Wholesale
Weekly Price
Exhibition
News
-
Trade Service
In the field of medicine, AI recognition "cough sound" has been used for a number of diseases such as pneumonia, asthma, Alzheimer's disease, these diseases will lead to the deterioration of the body's function, such as reduced vocal tract, respiratory function, etc. , although the human ear is difficult to identify, but through machine learning and signal processing, AI can recognize the difference, and more and more evidence that patients infected with the new coronavirus will have similar changes in body function.
the AI model developed by
MIT researchers is based on the AI model the team previously used to diagnose early-stage Alzheimer's patients, and after the outbreak, the team realized that it might be available to identify people with COVID-19 asymptomatic infections.
April, researchers set up a public cough data collection website that allows everyone to voluntarily submit cough recordings through devices such as web browsers, mobile phones or laptops.
, the site has collected more than 70,000 recordings, including about 200,000 cough audio samples, of which more than 2,500 were submitted by patients who have been diagnosed with new crowns.
AI model extracts the audio characteristics of cough records (Mel's frequency inverter coefficient) and inputs them into neural networks (courous neural networks, CNN) to learn about cough differences between patients with new coronavirus and healthy people.
specifically, the researchers trained three neural network models to extract vocal band strength characteristics, distinguish emotional states in speech, and learn from self-built cough data sets to identify changes in lung and respiratory function.
researchers selected 2,500 newly crowned cough audio samples and another 2,500 random cough audio samples, and used 4,000 of them to train AI models and test the accuracy of the model with the remaining 1,000.
test results show that the AI model is able to accurately identify new crown patients based on biological characteristics such as vocal band strength, mood, lung and respiratory function.
the model was 98.5 percent accurate when listening to recordings of confirmed cases and 100 percent accurate when listening to recordings of coughs in asymptomatic patients.
and the specificity of the model has reached 83% and 94%, respectively, there are no large number of false positive or false negative reports.
"We think this suggests that when you have new coronary pneumonia, the way people produce sound changes, even if it's asymptomatic."
" said Brian Subirana, a research scientist at MIT's Auto-Identification Laboratory, adding, "While the system excels at monitoring unhealthy coughs, it's not easy to see it as an authoritative tool for diagnosis."
" is currently working with a number of hospitals to create a more diverse database.
same time, the team is applying to the FDA to integrate it into the app, and if put into use, mobile phone users can simply log on to the app and cough twice to find out if they may have contracted the new coronavirus.