Indian Scientists Devise Low-Cost, AI Based Software to Detect Covid-19

NEW DELHI: As India works hard to procure testing kits from abroad, there might be something worth considering from Japan to make up for the shortage.
A team of Indian medical scientists in Kyoto, with support from a professor and students from IIT Roorkee, has developed an AI based software which uses X-ray to not just detect cases of the novel coronavirus but also predict the same in asymptomatic patients.

The software is said to work with 99.69 per cent accuracy (though from a small available sample size) and is expected to get better with the availability of more Covid-19 images. The Osaka-Kobe consulate general of India supported the friendly cooperation for research on the issue.
As the software is based on just X-ray, invasive swabs can be avoided and rapid screening (100 images in 3.63 seconds) facilitated. This would also allow authorities to do away with the need for keeping patients in quarantine for long.

“No touching is involved and hence there’s no need for protective equipment. It’s also more economical with no additional cost/ consumables and can be easily extended to cover any new mutated viral strain. It can also distinguish other clinical complications like pneumonia,” says Kyoto based medical scientist Namasivayam Ganesh Pandian, who led the research along with his student Vinodh Joseph and IIT Roorkee professor Balasubramanian Raman.
The software design is a 2-step module in which input image is processed sequentially in two streams that can not only answer Yes/No for COVID but also can distinguish false positive non-Covid cases like pneumonia. The detection is said to be at a level better than the human eye. While the possibility of a large sample size cutting into accuracy initially is not ruled out, it is ultimately expected to remain in the range of 99.69 to 100 per cent.
“The approach of this system is very simple. From the real time perspective, we take the chest x-ray image of the patient and send it to the Deep and Machine Learning Model, which in fraction of a second can determine whether the patient is affected with COVID-19 or Pneumonia or is Healthy,’’ adds Pandian.
The scientists and researchers working on the software are currently getting data for analyses from University of Cyberjaya in Malaysia. As to when the software will be available in India, at least 3 state governments are already said to be in touch with the team working on it in Kyoto.
Pandian says the software will be made available free of cost and that his team is looking for requests for on-field validation. The program though needs more data/meta-data to become more intelligent. At this moment, as he says, this can be seen as a surveillance tool that can aid authorities to predict and decide who should be tested for polymerise chain reaction (PCR) for confirmation.

“We need large-scale analysis with randomized patient data for validation and future application,’’ he says.
On the use of an AI based software, he says there is a consistent improvement in accuracy with incremental data and that it will ease the burden related to pathology tests and workforce shortage. “It can perform even in resource-constrained settings like villages and quantify disease progress and recovery real-time,” he adds.

Source: ToI

Image Courtesy:Imaging technology news