Full Press Release Details
R A T E P A R T I C I P A N T S
Dr. Howard G. Berger, President
and Chief Executive Officer, RadNet Inc.
Sorensen, Chief Executive Officer and Co-Founder, DeepHealth Inc.
R E N C E C A L L P A R T I C I P A N T S
Mitra Ramgopal, Sidoti &
Good day everyone. Welcome to the RadNet
Inc. DeepHealth Artificial Intelligence Call. Today's conference is being recorded.
At this time, I would like to turn the
conference over to Dr. Howard Berger, President and Chief Executive Officer of RadNet Inc. Please go ahead, sir.
Thank you, Operator. Good morning everyone.
This is Dr. Howard Berger calling in. Two
days ago RadNet announced that it had acquired DeepHealth and formed a division for artificial intelligence that will be led by
the president of DeepHealth, Dr. Greg Sorensen. Today, we want to talk about RadNet's decision to move more substantively
into artificial intelligence as well as the strategy that hopefully RadNet and DeepHealth will be executing together to accomplish
Ten years ago RadNet made a critical decision
to invest and own its own information technology platform, which we call eRAD, which would serve as the backbone for managing a
widely distributed network of imaging centers. The premise of this decision was based on both economic and operating requirements,
(inaudible) representing an existential necessity. Indeed, the strategy has proven successful as operating RadNet without this
capability seems improbable.
I believe RadNet, and in fact all of radiology
and imaging is at a similar crossroads. Artificial intelligence will undoubtedly impact every facet on how the practice of imaging
Given the size to which RadNet has grown,
and likely to continue growing, artificial intelligence becomes another existential necessity to maximize operational and clinical
Over the past year RadNet and DeepHealth
have been corroborating in the use of artificial intelligence for mammography. During this period we came to know the DeepHealth
team, and in particular, Dr. Gregory Sorensen. When RadNet made the decision to make a substantial investment into artificial intelligence,
it was obvious that Dr. Sorensen's credentials were uniquely qualified to lead this effort. Clinical and research radiologist,
business and technology experience, running the North American division of Siemens Healthcare, and evolving the DeepHealth effort
in mammography AI over the past four years.
The opportunities which lie ahead in artificial
intelligence practically fall into three categories: one, clinical accuracy and productivity enhancement; two, business and operating
efficiencies; and three, revenue enhancement.
Before I turn the call over to Dr. Sorensen,
I would like to give some perspective to the category of revenue enhancement.
Procedures performed in diagnostic imaging
centers generally require a prescription for the particular exam from the reporting physician. In addition, for advanced imaging
procedures, authorization is also required before the exam can be performed. The only exception to this rule is screening mammography,
which the patient can self-refer for annual or bi-annual follow-up. As a result of this capability, RadNet has sought tools to
improve compliance and increase volume by direct outreach to the patient population. Since RadNet implemented the White Rabbit
AI application, mammography volume in most of our regions has increased substantially. The benefit of using mammography and artificial
intelligence for more accurate and earlier diagnosis of breast cancer is rapidly being validated.
Through the recent technology advances,
the opportunity to use AI for prostate, lung and colon cancer screening are now possible. Similar to mammography, the potentially
new screening tools should eventually be adopted by patients and payors as both affordable and necessary, and as a part of population
health initiatives which improve outcomes and reduce costs. Thus, the use of AI for these new screening tools should create a significant
benefit which will help direct patients to centers which utilize this capability. Both RadNet and DeepHealth share this vision
and made the combination and collaboration opportunities compelling.
I'd like to turn the call over now
to Dr. Gregory Sorensen who will further elaborate on both the circumstances which led RadNet and DeepHealth to come together,
as well as his vision of the opportunities for the future in artificial intelligence. Greg?
Dr. Gregory Sorensen
Thank you, Howard. Hello everyone. It's
a pleasure to be with you on the call today.
I'd like to begin by affirming what
Dr. Berger has said about the potential impact for artificial intelligence, and in particular a form
of machine learning known as deep learning, to improve the health of women and men.
Dr. Berger's categorization of opportunities,
that is those three opportunities he mentioned-clinical enhancements, business operations and
revenue enhancements are also spot on. Before diving into the specifics of how DeepHealth's team and technology will address
each of these issues, I'd actually like to review a couple of commonly discussed concepts to help set the stage.
Let's start with an analogy that's
gotten a lot of attention, namely the phrase data is the new oil. In some ways, that's a very apt analogy, but I'd
like to say that a better one is that data is the new tight oil or tight natural gas. Getting the value out of the source requires
a lot of work, perhaps not precisely analogous to fracking but still a very energy intensive, sophisticated and time-consuming
process. If we were to continue that analogy, DeepHealth would be like an oil extraction company and RadNet would be like the Permian
Basin, a vast resource with tremendous potential.
Bringing these two together enables so
much opportunity. This analogy may be even more apt because it's clear that the two companies are such a good fit for each
other. From DeepHealth's perspective, it's far more efficient for us to be inside the sources of data that we need
to do with the AI, that we want to do, to seamlessly deal with collecting the data. Something that, of course, is made much easier
at RadNet, since you've heard RadNet owns the IT infrastructure itself.
We also both have instance access to friendly
and cooperative medical experts to ensure that our AI isn't developed in a vacuum. This is something that other imaging companies
struggle with from time to time.
From RadNet's perspective, having
what you might think of as the fracking company in the world show up and offer to help them realize the value of all this data
is also very exciting.
Bluntly put, I believe that the most valuable
assets that RadNet has are not currently listed on its balance sheet. The data and clinical scale that RadNet has is so tremendous
that the opportunity to unlock that value is just incredibly exciting. That's one concept I wanted to discuss.
Another concept that is quite common in
AI is to say that AI is going to replace radiologists. Well, I'm a radiologist myself and RadNet works with over 750 radiologists
directly, and I can tell you that there's never been a better time to be a radiologist. AI is not going to replace us. AI
We at DeepHealth and at RadNet expect that
our machine learning technologies will enable RadNet physicians, meaning our affiliated radiologists, to practice medicine at the
very top of their license. We at DeepHealth are building the AI to do the drudgery parts of the job, which will enable our physicians
to provide better quality care while also being more productive.
How are we going to do that? What are some
of the specifics? Well, let me open the hood a little bit on our technology and our team to help explore that.
As you know, at the core of DeepHealth,
from the name you might guess, is deep learning. Deep learning is the name given to a relatively new way that lets computers learn
to do certain tasks. In our case, the breakthrough is that we can develop deep learning technologies to actually help physicians
do a better job of interpreting images.
Deep learning consists of taking what is
called a model, which is really just a set of equations, linear algebra equations, and adjusting the coefficients and the variables
in those equations until the model produces the answers that you were hoping to see. The great thing is that this adjustment can
be done automatically by exposing the set of equations or the model to training examples. If you want your model to distinguish
pictures of dogs or pictures of cats, you could starts with a computer model consisting of lots of linear algebra equations, show
those equations thousands of pictures that you know are a bunch of dogs, and other thousands of pictures that you know are pictures