A New Artificial Intelligence Model Has Been Developed to Detect Covid–19 Disease From Cough Sound

COVID–19 and Artificial Intelligence (AI)

While scientists continue their fight against SARS–CoV–2, one of the deadliest viruses in the last ten years, with antigen diagnostic tests, tests that help diagnose and prognosis, drugs, and vaccine inventions, informatics mostly continue to work on early detection diagnosis, prognosis, and prediction in this period. The aim is to reveal systems with a low margin of error that can help the workload of healthcare professionals and early diagnosis and initiation of treatment. The most commonly used computer vision (automation of the processes of vision and perception in humans, high–level interpretation on digital images or videos on the computer) is the processing of radiological images. Automation of the same image interpretation process for the performance of many applications and imaging results can be easily accomplished with complex and powerful computational platforms of large–scale data such as Deep Learning. With deep understanding, the manual design of these features is eliminated, and a large amount of different classification and regression tasks are completed with higher accuracy.

Artificial neural networks were used to detect and monitor positive cases with chest radiography and computed tomography. Of these, conventional neural networks gradually learn patterns bypassing the inputs from the images through the computational layers.

These patterns can occur in a wide window from the edges, lines, and corners of the lung image to find their distinctive features and classify them accordingly.

It shows that most review studies on COVID–19 and Artificial Intelligence (AI) either focus on a single aspect of COVID–19 management or examine the same type of dataset (just like image processing).

The general approach used to incorporate Artificial Intelligence (AI) techniques that require clinical blood samples and radiography images to identify, classify, and diagnose COVID–19 is presented:

1. Various repositories have been created to store and share datasets related to COVID–19.
2. Besides data mining, different preprocessing techniques such as noise removal, data cleaning, feature extraction, partitioning, and feature analysis are mainly used to enhance the dataset and transform it into a more meaningful and effective representation.
3. Artificial Intelligence (AI)–based techniques and tools have been described using COVID–19 datasets to distinguish patients affected by COVID–19 from others.

Diagnostic Tools Based on Cough Waves and Respiratory Patterns

In addition to systems that detect and distinguish respiratory patterns such as tachypnea and different sound waves such as coughing, studies in which positive cases are detected using the evaluation of thermal videos formed by breathing are interpreted.

Do you suspect COVID–19 with every cough? Unfortunately, the only way to find out is to take a coronavirus test. However, scientists who found that the cough sound is different even in asymptomatic COVID–19 cases are working on a much easier method to remove this doubt.

With a new artificial intelligence model developed by a group of researchers at the Massachusetts Institute of Technology (MIT), it will be possible to identify COVID–19 patients who do not show symptoms.

The research conducted at the Massachusetts Institute of Technology (MIT) focused on the understanding that asymptomatic people, that is, asymptomatic people, experience some changes caused by the coronavirus. As a result, these people’s coughs are different from normal ones.

The researchers developed the model with tens of thousands of examples of coughing sounds and sounds made during speech. When the new cough sounds were loaded into the model, the Artificial Intelligence (AI) correctly identified the cough of those known to have COVID–19 patients with 98.5 percent accuracy.

Researchers who have developed an algorithm that can detect this difference, which the human ear cannot see, are now working on the model to be widely applied. For this, the researchers contacted a technology company and are working on developing a free mobile phone application.

With software developed with Artificial Intelligence (AI) in the USA, COVID–19 patients were detected only by the sound of their coughing.

The software diagnosed cases with 98.5 percent accuracy among those who tested positive for the coronavirus, while the success rate among those who did not show any other symptoms was 100 percent.

Experts say that the main difference in asymptomatic COVID–19 patients is that the human ear cannot hear their cough.

How was the model developed?

Before the outbreak, the research group worked on algorithms that detect coughing sounds recorded on mobile phones for accurate detection of pneumonia and asthma. In a side study, researchers recording forced cough sounds to see Alzheimer’s disease, which affects both nerves and muscles, tested whether algorithms could detect vocal cord weakness.

It was understood that the “mmmm” sound could indicate the weakness of the vocal cords in the model developed for this, which detects different degrees in the strength of the vocal cords.

In the second stage of the model, which was loaded with thousands of hours of speech, the algorithm was developed to analyze emotional states such as calm, happy, sad. Finally, the third stage focused on understanding the differences in the lungs and airways from the cough sound.

Researchers have focused their studies on this area since April, with the prediction that the model they are developing on the COVID–19 epidemic can be used to detect the disease.

Researchers who set up a website asked participants to record cough sounds and upload them to this site and fill out a questionnaire containing symptoms and test information about COVID–19 disease.

Researchers, who obtained more than 70,000 records, focused on the records of 2,500 COVID–19 patients, including asymptomatics.

Scientists emphasize that the detection of predominantly asymptomatic people is of great importance for controlling the epidemic. Since these people do not show symptoms, they do not get tested, and because they do not isolate themselves from society, they cause the circulation of the virus.