As Pakistan’s doctors, nurses and health staff are spearheading the battle with coronavirus, its students are committed to aid the process. A student duo at Ghulam Ishaq Khan Institute (GIKI) used their skills in artificial intelligence to create a detection software that can diagnose coronavirus patients.
The duo consists of Muhammad Aleem, a student of mechanical engineering, and Rahul Raj, a student of computer sciences. Both team members are final year students at GIKI with specializations in Computer Vision, Natural Language Processing, and Data Science.
Raj and Aleem collaborated to design an artificially intelligent software that uses the CT scans of patients to determine whether they have coronavirus or not.
Discussing his inspiration to design the software in the first place, Aleem told MIT Technology Review Pakistan that his interest in machine learning led him to the idea of an AI Detector. “When I got to know that China and South Korea have been using CT scans to diagnose patients, I realized that we don’t really have organized efforts like that in Pakistan.” That’s when Aleem got to work on the software. The first thing he did was form a team with his friend Raj, who was already working on other AI projects.
“I never meant for it to be on a huge scale initially, I was just doing it for fun, but I got a pretty good response and I saw how desperately in need of such technology our nation was,” Aleem reflected.
How it works
The AI detector can detect the onset of coronavirus symptoms and report their intensity and the location of the infection. According to Aleem, when a person gets infected with COVID-19, certain symptoms are manifested in the lungs, which can be detected from a CT scan. These are the same symptoms that cause difficulty in breathing for the patient.
The software application requires a .jpeg or .png form of an axial CT scan of the lungs to detect symptoms. It gives a new image as the output which pinpoints the symptoms and diagnosis of the patient. “Recently we ran some tests on it, and the detector accurately pinpointed the symptoms with 92% confidence,” said Aleem. While the software’s sensitivity toward the symptoms can be adjusted, it has the ability to diagnose patients who have minor symptoms as well.
The detection software is based on symptoms, but according to Aleem these particular symptoms occur in few diseases like pneumonia and coronavirus. So if certain symptoms are detected, it narrows down the cause of infection. “When we ran tests on this software, the true positives which indicate the accurate detection of symptoms, came out to be 100%. When differentiating between pneumonia and coronavirus, it resulted in 85% accuracy,” Aleem explained.
The team is working to improve this feature as well, taking aid from the experience of doctors and researchers in other countries. According to their research, the symptoms of coronavirus, contrary to pneumonia or any other disease, appear in distinct locations. They are using this information to improve the software which is still in the development phase so that it can detect coronavirus even more accurately.
Aleem noted that many friends, colleagues and independent doctors have helped them in understanding medical terms and procedures for the development of this detector. They also took aid from research and papers written by international experts to gain insight on the matter.
How it’s different
When asked how this software compares with other coronavirus testing methods, Aleem revealed that there are certain factors, like testing method, testing efficiency and required resources that have to be pinned down. “If you look at most testing kits, they need blood or swab samples. They also require manual power and take a lot of time. The only thing they are covering is the cost-effectiveness.”
While he supports all efforts to come up with new methods of testing, he believes that cost-effectiveness is not the only goal. “Our method doesn’t require much manual power, we don’t need a radiologist, and it takes much less time,” he said.
Talking about the cost of running the software, Aleem said, “We haven’t made any decisions regarding the pricing of this detector yet, because people of Pakistan are of higher priority. Right now we are focused on testing and deployment of the software.”
Even with no additional cost, the software would require CT scans, which can be expensive. This makes the detector costlier than common coronavirus testing methods. This may make it seem like the software is unfeasible, however, there are certain reasons the pair came up with this idea.
The first reason is the unavailability of testing kits in Pakistan. Across the country, hospitals are flooded with suspected coronavirus cases, but unable to test all of them due to a lack of testing kits. In this situation, the software can prove useful as it only requires a CT scan of the patient.
The second reason Aleem and Raj think this software can aid in diagnosis is that the ratio of doctors over patients is very small. Aleem argued that, “each doctor is treating a lot of patients. So we need a system, where the coronavirus can be detected automatically in a patient. With this software, we wouldn’t need to make new kits for coronavirus, we only need the CT scans.” He admitted that not all hospitals have the facility of CT Scans, but said that some hospitals in big cities do, and that could suffice.
Providing the third rationale, he claimed that it can take up to 30 minutes to test one patient for coronavirus using the standard kits. “If we have a lot of patients – and the number of patients in Pakistan is rising – how many patients can we really expect to test within a day?” The software, on the other hand, takes hardly five seconds to compute conclusive results whether a patient has coronavirus or not.
After the initial detection of coronavirus, doctors need to check the patient’s day by day progress. Aleem suggested that the software can help in doing that, as it can monitor the increase or decrease in the symptoms as well.
The technological resources required to run the software include a standard computer and a CT scan. However, the software needs to be tested with Pakistani patients’ data before it is ready to be used.
To test this software, the team used a small dataset as compared to what other countries have used to test similar technologies. They claimed that it was because developers in other countries have government support and access to a lot of data. “Currently, it’s just the two of us, and the data we got after a lot of struggle is very low in numbers.” Even though the results provided by the AI detector have been accurate with data acquired from other countries, they are still in need of Pakistani data to test the software. “Our next step is to get access to large amounts of Pakistani data. We would like to highlight the software through social media and catch the government’s attention so that they can provide us with the data,” said Aleem.
Aleem and Raj need data that consists of CT scans of Pakistani patients affected by COVID-19. They have created an email address, firstname.lastname@example.org, where the relevant data can be sent. The team is also working on a feature that will be able to comment on the current condition of the patient, specifying the severity and working conditions of the lungs.
The creators are hopeful that their software can work on a mass level in Pakistan because similar software has already been put to work in other countries. “In China, there are thousands of CT scans happening every day. They have saved a lot of money and a lot of manpower this way. I believe that if we use this software we could do that as well,” Aleem said.
As the increase in number of coronavirus patients in Pakistan has been exponentially higher than that in China, they fear that the doctors and health staff might get overwhelmed soon and hope that their software can stop that from happening. China and South Korea have also implemented techniques that use CT scans to detect coronavirus in patients, however, they haven’t open-sourced their software, so it is hard to say how similar it is to the AI Detector developed by Aleem and Raj. This software is the first of its kind developed in Pakistan.