Summary
With Covid-19 pandemic raging all around the world causing
harm to public health and livelihoods, this is a good time to strategize how AIcan be used effectively to help the health care industry to fight this pandemic
in particular and overall preparedness in general. The 3 main areas that we
believe AI can help are,
·
Automated monitoring of mask and other Personal
Protective Equipment (PPE) compliance to prevent the spread of infections.
·
Data aggregation and building predictive models
to identify potential hot spots for the spread of infection.
·
Building AI based tools to act as a physician’s
aid in diagnosis and treatment of Covid-19 positive patients.
Each of these three use-cases are described in detail below
1.
Automated monitoring of mask and other Personal
Protective Equipment (PPE)
Computer-vision based AI models can be trained to detect if
humans are wearing PPE like masks and if health care workers are wearing PPE
suits, gloves, face-shield etc., Experts all over the world concur that wearing
face masks is a simple yet important way to prevent spread of infection.
However, monitoring the PPE usage among general population is a vexing problem.
Lack of monitoring will lead to lack of PPE compliance and will risk an
emergence of the disease spread that the world has worked hard to contain so
far. Optisol Datalabs team are experts in building such AI models for PPE
monitoring and compliance in industrial space even before this pandemic took
over the world. In fact, we are one of the only AI service providers who have
successfully implemented PPE compliance monitoring and tracking in industries
to improve health and safety. We are re-training such models currently to
detect health-care related PPE like masks, gloves, safety suits and face
shields.
The model we train are optimized to run in edge devices like
cellphone enabling rapid deployment of such models without investing in
expensive hardware. Such cellphone turned AI devices can be programmed to send
safety violations to the concerned authorities to quickly take preventive
actions. The detailed process flow on how to train and deploy such AI models is
described in the diagram below. A demo of our Computer Vision based AI model
trained to detect mask, safety suit, face shield and gloves is published at this
link in our youtube channel
2.
Predictive models to identify potential
infection hot spots
Daily data on current Covid-19 positive cases are published
in various news outlets. It is possible to collate all these data and build a
model that can predict the likely hotspot of the infection to warn the
residents in those areas to exercise caution. Google and Apple are about to
publish updates to their platforms that allows monitoring of the movement of
the smart phone users in a way that does not violate the privacy of the users.
This will allow more granular collection of real-time data to build such
predictive models with increased accuracy and granularity. Such models are
effective tools for public health organizations and individuals to get access
to accurate data and to prescribe the least intrusive policies that allows
resumption of economic activity at the same time reduce the risk of infection
spread. We have extensive experience in building such time-based predictive
models. We are currently working on build a model from the publicly available
data right now.
 |
| Computer Vision Analytics |
3.
Models to act as physician’s aid in diagnosis
and treatment
Another important use of AI in health care industry is to
build models that act as physician’s aid in disease diagnosis and treatment.
There was a famous experiment conducted in the US where the AI can predict
heart attack and stroke better than trained physicians (https://time.com/5784090/ai-heart-attack-stroke/).
AI models can operate on large number of parameters that even trained
physicians cannot always reliably compute to make their decisions. In such
situations, AI becomes a valuable tool at the physician’s disposal for to
increase accuracy of the diagnosis and effectiveness of the care. There has
been lot of interest in just using chest x-rays to predict Covid-19 and there
are a lot of service providers who claims to build an accurate diagnosis tool
to detect Covid-19 from a chest x-ray. Such claims are not scientifically
validated, hence cannot be relied up as effective diagnostic tools. Such hype
in fact causes skepticism in using AI in health care amongst health care
providers. This is quite unfortunate. Because AI is proven to be an effected
diagnostic aid if used correctly. Chest x-rays are but one data point in the
diagnosis of the disease. There are other effective data points like travel
history, place of residency, potential contacts (neighbors, family members,
colleagues) with a positive diagnosis, other symptoms being experienced like
fever, cough etc., staying in a nuclear family with mixed age group members,
past health history of pulmonary diseases etc., When all these data points are combined
and fed to a AI model, we will greatly increase the predictive accuracy of the
tool. In fact, these data can be collected from the patient at their homes and
they do not even have to visit a hospital and potentially risk spreading the
infection.
Summary
These 3 use cases span infection control, monitoring the
spread of infection for effective public policies and diagnosis and treatment
of the disease. In our opinion, AI is somewhat under-utilized in how it can be
effectively used to help mankind traverse the current pandemic successfully
with the minimum loss of life and property. We are in the fore front of developing and deploying these AI models.
