> A new way to apply audio detection diagnostics
On different occasions we have talked about how new technologies are setting a new course in how we conceive health and how medical care relates to the patient. On this occasion, a new development of artificial intelligence that allows the detection of positive cases of Covid-19 through the audio of the person coughing made the news. The development and implementation was carried out in the public sector, specifically by the Government of the City of Buenos Aires, in Argentina.
Currently, the Government of the City of Buenos Aires maintains a messaging system through Whatsapp that has stood out for its conversational experience. "Boti" is the name of the character that contacts the citizens of Buenos Aires to provide information about the application of vaccines, Covid-19 detection operations and other essential information that concerns the civil society of that city. From this platform, the pilot test was carried out prior to the general launching of the new way of detecting positive cases.
For the pilot test, a selection of cases of people awaiting PCR test results was made. Once they had been sent, they were consulted about the possibility of sending an audio, following specific guidelines for recording coughs. The bot was trained with a bank of 140,000 coughs, said Felipe Miguel, Chief of Cabinets of the GCBA.
How does IATos work?
The procedure is simplified in three steps: the person can record and send an audio cough. Once the audio is received, IATos analyzes the sound: if it matches the positive case patterns, it recommends the person to take the COVID-19 test.
1 > Audio recording : The instructions indicate the playback of 3 coughs per subject. Once the voice note is received, the artificial intelligence system will analyze the sound: if it matches the patterns of positive cases, it will recommend the citizen to be tested for coronavirus. The collection of audio files was approved by the Ethics Committee of the Elizalde Hospital.
2 > Neural network analysis: A deep learning based cough detector associated with COVID-19 is used for the analysis. This involves a three-class classifier based on convolutional neural networks using Mel spectrograms as input information. Due to the subtle differences that exist between cough images corresponding to healthy patients from those with various respiratory diseases or with COVID-19, a neural network with more layers was required. For this, a CNN architecture was used .
3 > COVID-19 detection: At the moment, with limited data available, the overall accuracy of the classifier based on deep transfer learning is 88.00%. However, from GCBA, it is stated that future work will continue to improve this model as more training data becomes available.
The use of audio generated by people as a biomarker for various diseases opens up a wide range of ways to reach an early diagnosis. In addition, this methodology offers the possibility of being massively applied through platforms with the use of cellular devices. Technology accompanies and it is a key moment to be able to imagine a not so distant future in which medical care can start from the palm of the patient's hand and accompany him throughout his life in an active and effective way.
At SkyMed we specialize in bringing technology and health closer to enhance the processes of your institution. Let's talk? Leave us your details here to arrange a virtual meeting with one of our representatives.
Remember that to learn more and stay up to date with the latest news from the world of technology and business, you can subscribe to our newsletter by clicking here.
