I appreciate your analysis and feedback, in the rage to diagnose the COVID, it was necessary & vital thereof, to divert the attention of the Machine Learning experts and all the relevant individuals to pursue them and to aid them with some ideas on which the further research can be pursued. Moreover, I acknowledge it is not the sole responsibility of Data Scientist to push them into this idea, rather it is a teamwork thru which physicians & ML experts both should have to work together to reach to some concrete conclusion.And I believe once we successfully got this, we'll be able to develop its drug/treatment soon after.
------------------------------
Balaji Karumanchi
Sr Manager
Natsoft Corporation
------------------------------
Original Message:
Sent: Jul 02, 2020 09:59:11 AM
From: Mark Yanalitis
Subject: The Role of AI In Mitigating COVID-19 Confusion
I read this through pretty quick. I will take a slower read. By way of career history my graduate work is in Human Physiology primarily in cardiovascular medicine and some experience in transplantation (heart, liver, kidney). I spend a decade or more in applied research working in major clinical trials. I change careers and now I work in big data.
The copy makes too big of a leap right here and it damps the persuasive impact,
"Perhaps most significantly involving COVID-19, as primarily a respiratory-related malady, experts have employed deep learning algorithms for detection of COVID using images of CT scans."...
By majority the people presenting in ER's are at an advance stage of lung insult. You may be under a word limit, but I think you need to pull back and show how AI and ML image analysis is distinguishing between COVID aveolar damage, and that of damage by other sources. Vape Popcorn Lung and Bacteria Pneumonia would present exactly the same as COVID until H&P and differential diagnosis ruled other sources of insult. The AI and ML value seems to be opportunities to increase diagnosis sensitivity such as in the case of early detection and correlation between asymptomatic patients CT scans and their blood test panel lab values for example. Currently, chest CT scans (even deep "lawnmower" scans) will report patchy abnormalities of damage but a confirming differential diagnosis is needed using overlaying Ultrasonography on top of the CT discoveries to be sure of the extent of COVID diagnosis. Right now the statistical correlation is very good between both image exams, but perhaps AI and ML can make positive connections between chest CT scans and overlaying ultrasonography earlier in the disease course (e.g. seeing things that physicians cannot with the naked eye and in OLED screen). Right now the AI and ML clinical image data set available for training is small generally speaking - its getting better - but data wrangling in the clinical setting is a big issue for a number reasons leading to training sets that lack richness, depth, and dimensionality.
In the end analysis I might expressing a bias as a former applied research professional; take that into account. The paper spins off the DCNN/NN discussion of the article without setting up a clear clinical example where the techniques contribute to increased diagnosis sensitivity repeatably and reliability.
Good stuff. It's hard work. Hope this helps.
Mark Y.
------------------------------
Mark Yanalitis
Original Message:
Sent: Jul 02, 2020 09:14:06 AM
From: Balaji Karumanchi
Subject: The Role of AI In Mitigating COVID-19 Confusion
My Article published on "Health-IT Outcomes" trade publication. I appreciate if you all provide the feedback.
https://www.healthitoutcomes.com/doc/the-role-of-ai-in-mitigating-covid-confusion-0001
------------------------------
Balaji Karumanchi
------------------------------