Reshma Suresh heads Operations at Qure.ai. She has over 8 years of experience in building and scaling product innovations in healthcare. She focuses on strategic operations and is passionate about new technologies and last mile healthcare delivery. With a background in engineering as well as management, she works closely with clients and the product specialists to convert products to meaningful solutions.
She strongly believes that true impact can stem only from a patient-centric healthcare approach.
Summary
In this episode, Reshma starts with the early days of Qure.ai and the foundational aspects that are vital. Then she leads us through the challenges the pandemic presented and how Qure.ai was able to pivot their solution towards screening for Covid-19.
We then discuss ML-Ops that is critical to engineering delivery and Reshma shares examples on how important it is for the ML pipeline to support various solution needs i.e. speed, accuracy, reliability and reproducibility.
Audience Questions
[00:36:30] What challenges in the AI in Healthcare are not much researched and why is that so? Is that because of the lack of relevant data?
Aravind Naidu — Machine Learning Engineer @ Super Simple Software, India
[00:38:00] What is one big challenge they thought would be difficult to overcome that was easily solved and what was one easy thing they thought that turned out be a difficult problem?
Umashankar S — Manager R & D @ Hewlett Packard Enterprise, India
[00:41:50] The success of AI is fully dependent on access to large and diverse set of data. How are they integrating with hospitals to have access to that data to output “information” of value?
Kapil Assudani — Chief Information Security Officer at Edwards Life Sciences, USA
Topical Questions
[00:08:53]: You talked about backing your results with research or maybe publications in peer reviewed journals or conferences. How important has that been to get the sort of acceptance that your solution is at par or better?
[00:15:53]: If your solution failed to identify somebody who actually had COVID, then that person potentially goes back into society and continues to be an asymptomatic carrier. So did you have to think hard on how you could minimize that?
[00:18:10]: Tell us what is your definition of ML Ops or AI Ops and why do you think this is important in your scenario?
[00:23:43]: …there are several considerations typically, your entire ML pipeline end to end, that is your start to end or input to output could be designed for speed…it could be for accuracy, it could be for reliability, reproducibility right?…How would you pick, or in your organization or in your culture, where do you tend to focus more?
[00:30:42]: … what sort of mechanisms have you thought of, once your solutions are deployed to collect the relevant metrics, monitor for performance…?
Quotable Quotes
[00:09.40]: These publications, these validations, these comparisons either independently of your product or your product versus other products or with the gold standard or with the existing standard of care practices is very, very important to establish trust or reliability because we are dealing in healthcare.
[00:21:20] … so instead of dictating your end-user, you should instead have your technology or product to be compatible, to handle all these possible scenarios from the field that could go wrong.
[00:26:45]: If I am a patient and I go to hospital A versus hospital B even though my X-rays are taken by two separate technicians trained in different settings. … the final output should be exactly the same if these two hospitals have our software; both of these should predict the exact same for an individual who’s taking an X-ray, if the X-ray or CT is taken correctly.
[00:31:55] … positive is something which you need to probably keep working on and negative is something which you need to keep solving for.
Notable Mentions
[00:03:10] Mumbai, India
[00:06:10] STOP TB — A worldwide partnership that is leading the way to a world without tuberculosis (TB).
[00:22:45] ML-Ops — An engineering culture and practice that attempts to unify ML system development and ML system operations (from a Google paper).
Pointers to past You+AI Podcast episodes
[00:28:28] S2-E8 Leading the way with Digital Therapeutics, with Hicham Naim
Connect
LinkedIn: Reshma Suresh
Twitter: @_sreshma
The You+AI Vodcast
A companion video segment full of fun and candid moments.
Check it out here!