Page 37 - AIH-2-1

Artificial Intelligence in Health Deep learning on chest X-ray and CT for COVID-19

computer-based automated tools can be extremely useful, few confusing cases need to be cross-examined by trained
particularly when faced with a logistics bottleneck or medical professionals in another country, say “B,” who can
when the entire medical infrastructure is overwhelmed for undertake the post-ML decision-making with the help
some reason or other, which can be very handy during the of well-developed resources natively available in “B.” It is
outbreak of some other potent infectious diseases. important to note that the present protocol is performed

The method outlined in the present article could be a by an ML algorithm implemented on a computer, and
very important building block in an emerging integrated the actual job of a medical expert is minimal, limited to
digital medical doctrine that can ensure a more reliable, only overviewing/double-checking the assessment of the
personalized, and targeted medical intervention even computer to eliminate the remote possibility of mistakes
within the ambit of a very large-scale pandemic control (since the machine accuracy is more than 94% even
initiative spanning across prefectures/territories/states/ without human intervention, as demonstrated in a later
countries as previously outlined. One of the foundational section). Such global alliances can be a game changer in
2
principles of modern-era scientific practices is to decouple providing equitable diagnostics to underdeveloped regions
critical processes to maximize control over them and of the world (Global South), as already demonstrated in the
create provisions for more efficient resource allocation. case of vaccination under the COVID-19 Vaccines Global
12
In the RT-PCR test, the processes of physical examination Access, (COVAX). Moreover, free exchange of valuable
of the patient by the doctor, sample collection, and actual medical data across the boundaries will have the potential
evaluation of the testing results are highly integrated and to create huge synergistic effects, for example, in faster
codependent. A decoupling approach in this scenario is identification of newer strains, proper epidemiological
difficult to achieve. In the present ML-based method, the analysis and consequent prevention strategy, and most
physical examination of the patient in the form of taking importantly, a reliable prediction about the future
an X-ray radiograph and the evaluation of the testing trends based on hard data obtained through data-driven
results from the ML model can be effectively decoupled, approach. Developing an automated analysis system can
thus offering a modular approach that tremendously therefore save valuable time for the medical professionals
enhances flexibility in operational and clinical protocols in the country “A,” which could be best utilized to address
for pandemic control. This effective decoupling and some other critical conditions. This will also ensure
modular approach will be a crucial component for the optimal resource utilization and advanced preparation for
network-centric digital ‘health stack’ for the future the pandemic in country “B.” As pointed out earlier, the
pandemic control by effectively predicting its geographical detailed digital architecture and infrastructure planning
occurrences and this will be dealt with in the second will be presented in the follow-up article. It is important
article (in writing progress) in this series. Let us illustrate to note that such a design will not be COVID-19-specific
this point further with a simple example for the sake of but flexible enough to handle a broad spectrum of other
brevity (the comprehensive architecture will be outlined diseases.
in the follow-up article): say an economically weaker COVID-19 radiograph (CORAD) scores from CT
country “A” lacks enough resource pool of highly trained scans are considered a definite diagnosis of COVID-19
medical professionals and/or RT-PCR test kits reserves. even in the case of negative RT-PCR results. Chest X-ray
However, medical X-ray radiography is one of the oldest or CT scan of a COVID-19-infected patient generally
and most common diagnostic techniques. As a result, reports abnormal findings, such as ground glass opacities,
most countries (including “A”) are already well-equipped and coarse horizontal linear opacities scattered throughout
with X-ray devices. The approach proposed in this article the lungs, often with consolidation. They represent
13
requires minimal training of personnel who can supervise tiny air sacs getting filled with fluid. Another finding is
X-ray radiography tests satisfactorily (that is, just to take called the “crazy paving” pattern, which is caused due to
a satisfactory radiograph; in the case it is not satisfactory, swelling of the interstitial septum along the walls of the
it can be automatically flagged by a computer/professional lung lobes superimposed on the background of ground
located in another country to re-do the test). However, the glass opacities. The latter finding is observed in the
difficult part of this approach is to train a large number advanced stage of infection. Although RT-PCR remains
14
of medical professionals in analyzing/getting familiar with the gold standard procedure for COVID-19 detection,
X-ray images for COVID-19 or future pandemic detection. these findings in X-ray images can help in the initial
Here, the great advantage of decoupling the critical process screening of the suspected patients. There are six patterns
becomes clear. By using internet-enabled technologies, indicative of COVID-19 infection that can be observed
the tests and results can then be transmitted online for in a patient’s X-ray report. They are: (i) reverse batwing,
further evaluation by automated ML algorithms. Only a (ii) multifocal lower lobe predominant consolidation,

Volume 2 Issue 1 (2025) 31 doi: 10.36922/aih.2888