The COVID-19 pandemic looks like it’s not going to be the global event of 2020. With cases on the rise, reports of the virus mutating, the vaccine slowly rolling out, and lockdown measures back, COVID is going to be a big part of 2021 as well. This all means that the fight hasn’t stopped and we have to ramp up our efforts to combat this virus in any way we can. In this vein, there is a duo in Zimbabwe who working on a way to improve the COVID-19 diagnostic value chain through AI.
The first individual in the two-man team is someone that many of you may be already aware of. His name is Pardon Mukoyi and he previously worked on Cybersecurity system called EyeWatch. He has worked with organisations like the United Nations, AT&T, CERT-EU and many others. The other half of this team is a James Murombo who is currently studying Industrial and Manufacturing Engineering at NUST (National University of Science and Technology). He is also an artificial intelligence and deep learning enthusiast as well as having experience in Auto CAD (Computer-Aided Design), Solidworks, and Python Programming.
How did this all come about?
COVID-19 diagnostic tools have taken a great leap over the past year (ID-Now is an example of this). Researchers and medical health professionals have been working in tandem to streamline the diagnostic process as best as they can, without, of course, sacrificing accuracy.
3TB Desktop HDD
ACER Aspire 3
Linovo laptop think pad
Acer extensa 2519
One of the ways that both James and Pardon identified as a means of doing so was through AI and deep learning. The two arrived at this project after Kaggle made public a database of chest X-Ray radiographs that showed healthy patients vs those with COVID-19 affected patients.
The dataset that Kaggle gathered was through its Hack-D-Covid’20 which sought to build an algorithm to detect COVID-19 using chest X-Rays. This spurred them to develop a local solution through the data that was made available.
Another reason that motivated them to undertake this project was that RT-PCR (Reverse Transcription Polymerase Chain Reaction) although accurate, takes some time. There is a need for a local system that can help with an initial assessment of patients and after RT-PCR, monitoring the progression of the disease. One way of doing this more readily and quickly is by way of chest radiographs.
So how does the COVID-19 AI model work?
The prototype COVID-19 diagnostic system that Pardon and James is a CADx (Computer-Aided Diagnosis) system that incorporates Artificial Intelligence (AI) and machine learning. The pair are using Kaggle’s open-source data set to train to get their system to pick out peculiarities of COVID-19 infected patients vs healthy patients.
“By computer vision the model is able to visualize, learn, study patterns and extract features from these images using neural networks and will be able to predict new unknown images fed into the system giving a highly accurate diagnosis with an accuracy of 98.88%”
This accuracy is in the league of RT-PCR (98.2%) and ID-Now (85.2%), however, the two say that their system is meant to add to the laboratory tests rather than replace them in Zimbabwe.
So why is this important?
Looking at our situation in Zimbabwe a system like this one could be invaluable. For one thing, it could help healthcare professionals deal with the increasing number of patients. If this system reliably works they could be able to partially automate the initial assessment of patients.
This, however, won’t replace the trained eye of a radiologist or a physician, but it will make help them improve efficiency because they can test patients in bulk and this will guide laboratory researchers and scientists when they conduct further tests.
The CADx prototype, with help from local hospitals, could also aid in assessing other respiratory illnesses like Turbeculosis and Lung Cancer.
One big thing that stands in the way of this is how X-Rays are developed. Some institutions don’t have machines that can give out radiographs in JPEG or PNG formats. This will slow down the speed at which this system could work if it is deployed nationally.
Another issue, which is a little far off in the timeline, is integrating this system to X-Ray machines. If there is a way to get this in every hospital across the country then it could greatly improve the speed and efficiency of diagnosis and treatment.