Page 26 - Aesthetics&Dermatology_11_2018

RESEARCH REPORTS

AI HOLDS GREAT PROMISE

FOR VISUAL FIELDS LIKE

DERMATOLOGY, BUT FACES

MANY CHALLENGES

Finding robust data sources, correcting for variations in images and avoiding bias are all
hurdles to overcome.

Computer vision has great promise
for helping to democratize fields like
wound care, dermatology and more.
However, as companies explore this
potential, they’re also discovering a
number of challenges to overcome.
What most researchers and
The data problem developers want is an AI that
can look at pictures taken with a
“Getting the data is really the smartphone in a home environment.
biggest challenge, not the AI,”
Karen Panetta, IEEE fellow and
dean of graduate engineering
at Tufts University who studies
AI use cases in healthcare, told
MobiHealthNews in an interview
earlier this year. “We’ve already
got the models, we just need
more training data to validate this at Mount Sinai Medical Center in that’s problematic, right? … The
expertise. And then, again, getting New York, but she’s also working whole point of a classifier is to be
doctors to also validate, to get remotely with teledermatology generalizable across a population,
random things from a cellphone, company First Derm on improving and you’re limited to a small amount
and you want multiple doctors to do the company’s AI algorithms. She of data, that already is an issue
it because they have to agree.” says a traditional clinical research from a statistical standpoint.”
approach doesn’t bring in anything
There are existing clinical data near the scale needed for machine This is leading a number of
sets, but the pictures they contain learning. companies like First Derm to create
are clinical images, taken in direct-to-consumer teledermatology
controlled conditions, often with a “The issue in general is that a or dermatology triage tools, in which
dermascope, which is a specialized lot of the data people are pulling patients consent to sharing their
medical instrument for taking in from healthcare studies, the data in exchange for free access.
pictures of the skin. enrollment process for clinical trials
is generally very slow, and very Even companies like VisualDx,
Furthermore, getting access to manual. Not that this is specific to which have a robust dataset
medical datasets is very difficult, any one institution, but if you have from years in the CDS space,
since patients have to have sample sizes in the low hundreds have to balance patient privacy
consented to have their data used or not even 100 — and there are considerations.
in research and most have not. plenty of studies that are 30, 50,
Even if they have, the researchers 70 participants, just because it’s “In our professional tool, when a
have to secure IRB approval for so difficult to get a willing cohort doctor takes a picture of a patient
access to the images. that will show up for all the testing the image is analyzed on the phone
you need, and generally you do and the image is dumped,” CEO Art
Mary Sun is a medical student recruitment locally so it’s just the
at the Icahn School of Medicine patient population that’s available to 27
you and so on and so forth— and

26