With input from Dr Monica Vasudev

1116 IMA-CMAAO Webinar on “Artificial intelligence in #COVIDTIMES”

17^th October, 2020, 4-5pm

Participants: Dr KK Aggarwal, President CMAAO, Dr RV Asokan, Hony Secretary General IMA, Dr Ramesh K Datta, Hony Finance Secretary IMA, Dr Jayakrishnan Alapet, Dr GM Singh

Dr Brijendra Prakash, Dr Ranjan Wadhwa, Dr S Sharma

Faculty: Dr Vidur Mahajan, Associate Director, Mahajan Imaging & Head R&D, Centre for Advanced Research in Imaging, Neurosciences & Genomics, New Delhi

Key points from the discussion

• All of us use artificial intelligence in our day to day life. Some examples are Google maps, Amazon, Netflix search; keyboard tries to figure out what we intend to write and gives word suggestions, etc.
• Artificial intelligence is not a new concept; it existed even in the 1950s, although there was not much of computing power then. But it has recently become popular because of the exponential increase in computing power.
• AI is the ability of the computers to learn without being programmed, i.e., the computer figures it out for itself. Basically this involves using high end computing to find hidden patterns in data.
• There are three types of artificial intelligence: Artificial “narrow” intelligence, artificial “general” intelligence and artificial “super” intelligence.
• Artificial “narrow” intelligence is the ability of the machine to do one thing very well, e.g., the Deep Blue computer which beat Gary Kasparov in chess. This machine can only play chess. Currently, machines are artificially “narrowly” intelligent, i.e., intelligent in only one domain.
• Artificial “general” intelligence: learning one thing and then extrapolating that knowledge to another; humans are generally intelligent at varied aspects.
• Artificial “super” intelligence is a much debated topic and describes the point in time when AI becomes stronger than humans.
• AI applications can be broadly categorized into two: what humans can do but AI does better and faster and what humans cannot do.
• Supervised learning - machine figures out patterns in the data itself; most healthcare AI is focused on this, but it is labor intensive and involves feeding labelled data into algorithms so that patterns in the data are recognized.
• Unsupervised learning is what human beings are – we automatically learn from our environment and can transfer knowledge from one aspect to another. It is more computationally expensive and currently its applications in healthcare are limited. Lot of work is being done on can machines learn without being trained.
• Steps in creating an AI algorithm: Find data (heterogenous data related to the problem in question as much as possible), train the model, validate the model and test the model.
• COVID has driven adoption of digital health tools – telemedicine guidelines, eSanjeevani; most records are now getting digitized.
• AI is helping in the war against the entire COVID-19 landscape of prevention, diagnosis and treatment.
• Globally, AI is helping in many public health decisions, e.g. Aarogya Setu App, Delhi govt. worked with IIT Delhi to find out which areas to lockdown and which to open up using AI, proximity sensors for contact screening
• Machine learning plus genome sequencing helped Kerala to identify where their cases were coming from.
• AI tools are also an essential part of vaccine development. AI technology is being used to sift through large amount of data and find the most useful literature to make the ideal vaccine (CORD-19 – COVID-19 open research dataset, a free resource of more than 280,000 articles about COVID-19 for use by the global research community); molecular biology – AlphaFold (computational predictions of protein structures associated with COVID-19) – using machine learning algorithms to predict how protein structures can interact with each other and potentially vaccine can impact body. AI is being used in vaccine trial – who to give the vaccine to, how many doses, etc.
• Chest X-rays have low sensitivity for COVID diagnosis but is best used in high suspicion situations in resource constrained areas.
• A lot of AI work is going on in India. AI systems built for TB are being used for COVID by many people working in this area.
• Two applications of chest X-ray: Diagnosis and triaging and monitor disease progression.
• CT scan has better sensitivity in diagnosing and monitoring; Predible algorithm helps to analyze 15,000 CTs in a month; it is trained on 40,000 COVID images and has 95% volume concordance with calculation of volume involvement by a radiologist.
• In the absence of RT PCR, AI for COVID played a vital role.
• Many novel attempts at diagnosing COVID using AI are underway, e.g., trying to diagnose from cough samples (AI4COVID-19), detection of COVID-19 from routine blood examinations.
• AI is also at the core of drug development and drug repurposing (The Lancet, Sept. 18, 2020).
• Future of medicine is in AI.

Dr KK Aggarwal

President CMAAO, HCFI and Past National President IMA