Welcome to the Episode 345, part of the continuing series called “Behind the Scenes of the NetApp Tech ONTAP Podcast.”
This week, Kim Garriott (kim.garriott@netapp.com) drops by to teach us all about digital pathology and how it is the next wave of change in the healthcare IT industry.
For more information:
- Simple and Scalable Digital Pathology Analysis with NetApp AI
- AI and Digital Photography
- NetApp Medical Imaging
- etApp and Google Cloud unlock the power of medical imaging
Tech ONTAP Community
We also now have a presence on the NetApp Communities page. You can subscribe there to get emails when we have new episodes.
Finding the Podcast
You can find this week’s episode here:
I’ve also resurrected the YouTube playlist. You can find this week’s episode here:
You can also find the Tech ONTAP Podcast on:
I also recently got asked how to leverage RSS for the podcast. You can do that here:
http://feeds.soundcloud.com/users/soundcloud:users:164421460/sounds.rss
Transcription
The following transcript was generated using Adobe Premiere’s speech to text service and then further edited. As it is AI generated, YMMV.
Episode 345: NetApp and Digital Pathology
1
00:00:00,533 –> 00:00:02,168
This week on the Tech ONTAP Podcast.
2
00:00:02,168 –> 00:00:05,405
Kim Garriott joins us to talk to us all
about digital pathology.
3
00:00:07,040 –> 00:00:27,360
[Tech ONTAP podcast intro]
4
00:00:31,431 –> 00:00:34,100
Hello
and welcome to the Tech ONTAP Podcast.
5
00:00:34,134 –> 00:00:35,802
My name is Justin Parisi.
6
00:00:35,802 –> 00:00:37,170
I’m here in the basement of my house
7
00:00:37,170 –> 00:00:40,173
and with me today – on the phone –
we have Kim Garriott.
8
00:00:40,173 –> 00:00:43,009
So Kim, what do you do here at NetApp?
How do we reach you?
9
00:00:43,043 –> 00:00:43,977
Hi, Justin.
10
00:00:43,977 –> 00:00:47,414
I am the Chief Innovation
Officer of our NetApp healthcare
11
00:00:47,714 –> 00:00:52,152
and life sciences practice and also
the general manager for medical imaging.
12
00:00:52,152 –> 00:00:57,757
And you can reach me
at kim.garriott@netapp.com.
13
00:00:58,258 –> 00:00:59,492
All right. Excellent.
14
00:00:59,492 –> 00:01:01,661
So Kim is here today.
15
00:01:01,661 –> 00:01:03,396
She’s been on the podcast
before to talk about
16
00:01:03,396 –> 00:01:05,732
healthcare and medical imaging
and that sort of thing.
17
00:01:05,732 –> 00:01:07,700
Today,
we’re going to talk about something new,
18
00:01:07,700 –> 00:01:09,736
something that’s
kind of cutting edge in the industry,
19
00:01:09,736 –> 00:01:12,972
and that is a concept known as digital
pathology.
20
00:01:13,373 –> 00:01:16,443
So, Kim,
could you please tell me what that is?
21
00:01:16,443 –> 00:01:17,410
Because I don’t know what that is.
22
00:01:17,410 –> 00:01:19,079
And I’m sure a lot of people
haven’t heard of this before
23
00:01:19,079 –> 00:01:21,081
other than, you know, what the words
mean, right?
24
00:01:21,247 –> 00:01:22,549
Yeah. No, exactly.
25
00:01:22,549 –> 00:01:27,654
It’s a really exciting field
that that is coming to life
26
00:01:27,654 –> 00:01:33,026
in healthcare and life sciences
and have some interaction with pathology.
27
00:01:33,026 –> 00:01:36,663
So any kind of tissue
removed from your body,
28
00:01:36,663 –> 00:01:40,533
say you had a gallbladder removed
or you went to the dermatologist
29
00:01:40,533 –> 00:01:44,170
because there’s a, you know,
a little mole or a bump on your arm.
30
00:01:44,170 –> 00:01:46,473
And they do a little biopsy sample
of that.
31
00:01:46,840 –> 00:01:49,609
That tissue or fluid from your body
32
00:01:50,009 –> 00:01:52,679
is set on a glass slide.
33
00:01:52,679 –> 00:01:56,049
And then that glass slide
is put under a microscope.
34
00:01:56,049 –> 00:02:00,787
And a medical professional – called
the pathologist – looks at the cells
35
00:02:00,787 –> 00:02:03,756
and the tissue on that slide
and makes a diagnosis.
36
00:02:03,790 –> 00:02:08,995
Most commonly that we can relate to
is whether that skin tissue
37
00:02:08,995 –> 00:02:12,465
is benign or malignant,
whether you have cancer
38
00:02:12,465 –> 00:02:15,802
or not, or the type of
cancer that you may have. So
39
00:02:16,803 –> 00:02:19,139
digital biology is
40
00:02:19,139 –> 00:02:23,042
really the next digital frontier
in clinical medicine.
41
00:02:23,042 –> 00:02:28,848
It may be the last major clinical workflow
to go through a digital transformation,
42
00:02:29,115 –> 00:02:31,851
and that’s when we take this glass slide
43
00:02:32,152 –> 00:02:34,854
and we digitize the slide
44
00:02:34,888 –> 00:02:38,791
so that the cells become an image,
45
00:02:39,526 –> 00:02:44,030
and then the pathologists can look at that
through their computer
46
00:02:44,164 –> 00:02:47,700
versus looking at the glass slide
through a microscope.
47
00:02:47,700 –> 00:02:50,236
So digital pathology,
48
00:02:50,970 –> 00:02:53,773
by definition, is a dynamic,
49
00:02:53,773 –> 00:02:58,011
image-based environment
that enables the acquisition,
50
00:02:58,011 –> 00:03:01,214
the management
and interpretation of pathology
51
00:03:01,748 –> 00:03:04,350
information that’s generated
52
00:03:04,350 –> 00:03:06,519
by digitizing that glass slide.
53
00:03:07,687 –> 00:03:09,455
So it seems like such a simple concept,
right?
54
00:03:09,455 –> 00:03:12,392
Like taking a glass slide
and turning it into a picture.
55
00:03:12,392 –> 00:03:14,594
So I can’t imagine it’s that simple.
56
00:03:14,594 –> 00:03:20,233
What is involved with digitally creating
an image of these slides?
57
00:03:20,233 –> 00:03:23,469
Is it as simple as scanning in a machine
or what else is involved?
58
00:03:23,469 –> 00:03:25,772
Is there any AI/ML impact there?
59
00:03:25,939 –> 00:03:28,841
It is a complex process.
60
00:03:29,142 –> 00:03:31,110
However, there are huge advantages over
61
00:03:31,110 –> 00:03:35,048
how we look at those glass
slides on a microscope today.
62
00:03:35,048 –> 00:03:39,018
So we would have the same process
of actually
63
00:03:39,018 –> 00:03:43,990
putting the tissue sample on the slide
and staining the slide or applying
64
00:03:43,990 –> 00:03:48,661
a chemical to the slide to make certain
cells be more visible than others.
65
00:03:49,062 –> 00:03:53,433
But then we will take that slide
and put them in trays.
66
00:03:53,433 –> 00:03:58,137
And these trays can hold anywhere
from one slide to hundreds of slides.
67
00:03:58,404 –> 00:04:02,942
And then you place them in a device
or a whole slide scanner,
68
00:04:02,942 –> 00:04:07,113
and then each slide is
then individually scanned.
69
00:04:07,113 –> 00:04:09,182
And there’s different ways
to scan the slides, depending
70
00:04:09,182 –> 00:04:14,187
on the type of sample that is on the slide
that you’re trying to digitize.
71
00:04:14,187 –> 00:04:18,057
So you can do tiling
where the scanner takes
72
00:04:18,057 –> 00:04:23,229
little tiny squares of images in a row,
or like they’ll do tile strips
73
00:04:23,229 –> 00:04:26,733
all at once,
where’s this one linear read of data.
74
00:04:27,033 –> 00:04:30,503
Or we can actually
for some types of tissue samples,
75
00:04:30,837 –> 00:04:35,008
we want to be able to have that X
and that Y plane scanned.
76
00:04:35,241 –> 00:04:39,779
But we also want to have the view
through the cells,
77
00:04:39,812 –> 00:04:43,683
as you can imagine, on a glass slide,
when you put the tissue on there,
78
00:04:44,417 –> 00:04:45,451
it’s not all flat.
79
00:04:45,451 –> 00:04:48,821
It’s not the cells
don’t line up all in one plane.
80
00:04:48,821 –> 00:04:49,188
Right.
81
00:04:49,188 –> 00:04:52,292
So there’s multiple planes or cells
82
00:04:52,292 –> 00:04:56,296
that can be stacked on each other
as you look down through a sample.
83
00:04:56,696 –> 00:05:01,267
So that type of Z stack scanning
and the intense scanning process
84
00:05:01,467 –> 00:05:06,239
gives us a really good depth
and allows us to see through that tissue.
85
00:05:06,239 –> 00:05:12,812
So once the slide is scanned,
then those images
86
00:05:12,812 –> 00:05:16,749
will then present on a view
or for a pathologist to look at.
87
00:05:16,816 –> 00:05:19,485
Now that’s the simplest workflow.
88
00:05:20,086 –> 00:05:24,624
Where the dynamics really come in
and where we can really enhance
89
00:05:24,624 –> 00:05:30,296
our ability to provide precision medicine
to patients, meaning individualized
90
00:05:30,630 –> 00:05:34,734
diagnosis based on their condition
and their sample alone
91
00:05:34,801 –> 00:05:37,437
is when we apply AI to this,
92
00:05:37,670 –> 00:05:42,709
we have a lot of improvements
that will come in this field
93
00:05:43,176 –> 00:05:47,380
because of the ability
to help the pathologist do their job.
94
00:05:47,380 –> 00:05:49,615
That’s kind of the first side of it.
95
00:05:49,615 –> 00:05:51,551
So what do I mean by that?
96
00:05:51,551 –> 00:05:54,987
Today when a pathologist has a glass slide
97
00:05:54,987 –> 00:05:59,025
and they put that slide
into the microscope, they have to
98
00:05:59,058 –> 00:06:02,762
first identify what’s the area of interest
that I want to look at.
99
00:06:02,862 –> 00:06:07,333
What is it on this slide sample
where the cells that I’m interested in
100
00:06:07,333 –> 00:06:09,502
are clustered
on? So first I have to identify that.
101
00:06:09,502 –> 00:06:14,941
And oftentimes they take just a marker
and they literally write on the glass
102
00:06:15,174 –> 00:06:17,410
and think about how small
those glass slides are. Right.
103
00:06:17,410 –> 00:06:22,181
You remember when you were in biology
or chemistry, how small those slides are.
104
00:06:22,482 –> 00:06:25,918
So a pathologist has to take a marker
and mark around that
105
00:06:25,918 –> 00:06:29,555
and then identify the cells of interest
and then count
106
00:06:30,089 –> 00:06:34,727
individual cells, which is very time
consuming for a pathologist.
107
00:06:34,727 –> 00:06:40,099
So at the front of the workflow,
we are working to develop AI models
108
00:06:40,466 –> 00:06:44,737
to mark the area or the region of interest
for the pathologist
109
00:06:44,737 –> 00:06:47,940
and began to do the cell segmentation –
or the cell counting.
110
00:06:48,307 –> 00:06:52,445
So the pathologists can really focus
on identifying the types of cells
111
00:06:52,445 –> 00:06:54,113
and the behavior of the cells
112
00:06:54,113 –> 00:06:59,252
versus just simply prepping the data,
if you will, for them to look at.
113
00:06:59,252 –> 00:07:04,590
And then beyond that, AI
and we have lots of AI developers,
114
00:07:04,590 –> 00:07:08,461
as you can imagine,
that are working on solutions or AI algos
115
00:07:08,461 –> 00:07:11,063
or algorithms to be able to help
116
00:07:11,764 –> 00:07:16,068
look at the cells, identify
the cells that are interesting
117
00:07:16,402 –> 00:07:21,441
and then compare those cells to databases
118
00:07:21,441 –> 00:07:24,877
of millions of other similar samples
119
00:07:25,311 –> 00:07:30,016
and be able to improve
that precision medicine diagnosis.
120
00:07:30,016 –> 00:07:34,854
For instance, today
we only have one approved AI algorithm
121
00:07:34,987 –> 00:07:41,461
in the field of pathology for disease
detection, and that is comparing
122
00:07:41,461 –> 00:07:47,099
tissue patterns from a needle
for biopsy used in prostate testing.
123
00:07:47,533 –> 00:07:51,037
And we’re able to take that needle
for a biopsy sample,
124
00:07:51,037 –> 00:07:53,706
compare it to a large database
of known cancers.
125
00:07:54,340 –> 00:07:57,109
And in the clinical trials for this
126
00:07:57,276 –> 00:08:00,913
AI model, they reduced the occurrence
of false negatives.
127
00:08:00,913 –> 00:08:04,650
That means I don’t
think I see any disease here.
128
00:08:06,152 –> 00:08:09,989
We have
reduced the false negatives by 70%.
129
00:08:10,323 –> 00:08:13,693
So that means 70% of the test cases
130
00:08:14,060 –> 00:08:18,264
that were used in these clinical trials
actually had cancer present.
131
00:08:18,498 –> 00:08:22,502
But the human pathologists by themselves
did not diagnose that.
132
00:08:22,502 –> 00:08:26,172
So 70% of the time
a patient had something questionable
133
00:08:26,339 –> 00:08:29,609
or wrong with them that we did not detect.
134
00:08:29,675 –> 00:08:35,681
So by applying AI,
we can be much smarter about our diagnosis
135
00:08:35,681 –> 00:08:38,985
because we’re comparing it
against millions of other tissue samples
136
00:08:39,218 –> 00:08:40,253
that may be similar.
137
00:08:40,253 –> 00:08:43,890
And in this example, we can see
how that really can impact lives.
138
00:08:43,890 –> 00:08:47,426
Now, I’m not saying
that 70% of false negatives
139
00:08:47,693 –> 00:08:51,564
that those people
had a critical condition.
140
00:08:51,931 –> 00:08:52,231
Right.
141
00:08:52,231 –> 00:08:52,999
But that could be
142
00:08:52,999 –> 00:08:57,870
someone that an early intervention
could keep them from becoming sick.
143
00:08:57,870 –> 00:09:00,640
Or maybe there’s other things
that we can do for that patient
144
00:09:00,973 –> 00:09:04,877
to help them have a great quality of life
and not actually be negatively impacted
145
00:09:04,877 –> 00:09:07,680
by the disease state
that may be occurring in their body.
146
00:09:07,880 –> 00:09:10,283
And that’s the beauty
of precision medicine.
147
00:09:10,283 –> 00:09:14,287
That’s the beauty of how we can apply
AI and make a difference in health care
148
00:09:14,287 –> 00:09:15,855
and in patient diagnosis.
149
00:09:15,855 –> 00:09:19,058
Yeah, another benefit of doing
this is – so that these slides I know
150
00:09:19,058 –> 00:09:22,795
if you’ve ever worked with any glass
slides and I remember in high school
151
00:09:22,795 –> 00:09:26,566
doing the biology stuff,
those things are real fragile, right?
152
00:09:26,566 –> 00:09:28,801
So they break, they break pretty easily.
153
00:09:28,801 –> 00:09:31,904
And then you have to think about
also the organic material itself.
154
00:09:31,904 –> 00:09:33,606
Over time it breaks down.
155
00:09:33,606 –> 00:09:36,842
So having images
taken of that really reduces
156
00:09:36,842 –> 00:09:40,713
the amount of worry about losing
this valuable data that you’ve had.
157
00:09:40,713 –> 00:09:43,382
You no longer are storing it
as a hard piece of evidence.
158
00:09:43,382 –> 00:09:47,620
You know, you’re using digital imaging,
which can last essentially forever.
159
00:09:47,687 –> 00:09:50,156
Yeah, that’s a great point, Justin.
160
00:09:50,156 –> 00:09:52,391
And that’s another huge benefit
161
00:09:53,159 –> 00:09:56,829
of this digitization of these whole slide
images.
162
00:09:57,196 –> 00:10:00,800
We use glass slides in a manner of ways.
163
00:10:00,800 –> 00:10:02,935
One is for primary diagnosis.
164
00:10:03,603 –> 00:10:05,972
The second is gaining a second opinion.
165
00:10:05,972 –> 00:10:06,272
Right.
166
00:10:06,272 –> 00:10:10,776
I want another pathologist
to look at this image or to look at this
167
00:10:10,776 –> 00:10:15,915
tissue to see if the diagnosis matches
the first pathologist that looked at it.
168
00:10:15,915 –> 00:10:18,951
And then we also have research
and education
169
00:10:19,318 –> 00:10:21,587
where these samples are very important.
170
00:10:21,587 –> 00:10:25,958
So as you can imagine, a lot of slides
get broken like you just alluded to.
171
00:10:25,958 –> 00:10:32,031
And by digitizing these images,
we are able to easily share the image
172
00:10:32,131 –> 00:10:36,502
where in the past, if we needed a second
opinion, we would have to box
173
00:10:36,502 –> 00:10:41,440
that last slide up and send it by courier
to another pathologist,
174
00:10:41,440 –> 00:10:45,344
which can take days,
which is a delay in diagnosis.
175
00:10:45,344 –> 00:10:47,313
And what if that slide gets broken?
176
00:10:47,313 –> 00:10:49,248
What if it gets misplaced?
177
00:10:49,248 –> 00:10:54,720
So not only the time consideration
of transferring that physical content
178
00:10:54,720 –> 00:10:59,225
to another provider, it’s
also the time to diagnosis.
179
00:10:59,225 –> 00:11:03,929
And if you are waiting on a cancer
diagnosis, days matter, right?
180
00:11:03,996 –> 00:11:04,830
We know that.
181
00:11:04,830 –> 00:11:10,369
So by being able to communicate
with a digital form of the tissue,
182
00:11:10,603 –> 00:11:15,675
then we are able to reduce the time
to diagnosis, be able to keep that sample.
183
00:11:15,875 –> 00:11:20,246
And then also if you look at use cases
with research and education
184
00:11:20,513 –> 00:11:23,816
as students are sitting in a classroom,
you want them all to be able
185
00:11:23,816 –> 00:11:26,886
to see the same interesting case.
186
00:11:27,653 –> 00:11:31,891
And today you have that pass
this glass slide around to students
187
00:11:31,891 –> 00:11:33,926
or hope that you had enough tissue samples
188
00:11:34,093 –> 00:11:36,595
that you could make a number of glass
slides out of it.
189
00:11:36,729 –> 00:11:40,800
But I think it’s pretty easy to see
in this way that if you have that tissue
190
00:11:40,800 –> 00:11:44,270
sample digitized,
then every student can look at that
191
00:11:44,270 –> 00:11:49,875
same sample at the same time and be able
to really improve the education process.
192
00:11:49,875 –> 00:11:52,712
And we don’t have to again
worry about breakage
193
00:11:52,978 –> 00:11:57,316
because once it breaks, then the tissue
samples to store it and no one can use it.
194
00:11:57,316 –> 00:12:02,521
The other thing about the glass slides
is that they are very heavy,
195
00:12:02,655 –> 00:12:08,594
if you can imagine that, and today,
organizations have to retain
196
00:12:08,928 –> 00:12:13,966
those physical glass slides for years
and years and years and that’s
197
00:12:13,966 –> 00:12:18,971
a lot of real estate to have to be able
to maintain these slides.
198
00:12:18,971 –> 00:12:23,375
And I heard a statistic
that Iron Mountain has kind of well
199
00:12:23,409 –> 00:12:28,848
over a billion of these glass slides
that organizations are having to retain,
200
00:12:29,281 –> 00:12:32,418
but they never go back and look at them
or very seldom, for
201
00:12:32,585 –> 00:12:35,488
primary diagnosis
do you go back and look at them.
202
00:12:35,488 –> 00:12:38,991
But they have to retain them
and it takes a lot of space.
203
00:12:38,991 –> 00:12:40,893
I’ll give you a couple of
interesting statistics.
204
00:12:40,893 –> 00:12:43,929
So there’s one large academic
medical center here in the U.S.
205
00:12:43,929 –> 00:12:48,801
that started scanning their glass
slides a few years ago.
206
00:12:49,201 –> 00:12:52,638
And they have scanned
almost 3 million slides.
207
00:12:52,638 –> 00:12:57,343
And just to put that in perspective
for you, a medium sized hospital
208
00:12:58,277 –> 00:13:01,313
probably has about 100,000
209
00:13:01,313 –> 00:13:04,083
pathology studies on their patients.
210
00:13:04,650 –> 00:13:07,753
So that would be 100,000 glass slides.
211
00:13:08,020 –> 00:13:12,558
This organization has scanned 3 million,
almost 3 million slides.
212
00:13:12,558 –> 00:13:16,629
That turned out to be around
three petabytes of data.
213
00:13:16,629 –> 00:13:20,166
But to the point of the weight of
the glass is 17
214
00:13:20,166 –> 00:13:23,869
tons of glass slides that were scanned.
215
00:13:24,670 –> 00:13:28,641
And they estimated that
if you were to stack those slides
216
00:13:28,641 –> 00:13:33,612
on top of each other, it would be 1155
217
00:13:33,946 –> 00:13:37,216
stories of slides, approximately.
218
00:13:37,216 –> 00:13:42,288
So that kind of just paints a picture
of the size of these glass slides.
219
00:13:42,288 –> 00:13:46,725
And if we’re able to digitize that,
that really takes a burden off
220
00:13:46,725 –> 00:13:50,229
of the health care organization
or off the life sciences organizations
221
00:13:50,229 –> 00:13:52,832
that may have to retain
that very heavy medium.
222
00:13:53,232 –> 00:13:56,702
Yeah, sounds like slides are kind of
like the tape of the health care
223
00:13:56,702 –> 00:13:57,303
industry, right?
224
00:13:57,303 –> 00:14:00,105
Like, trying to get away from that. Right.
225
00:14:00,105 –> 00:14:03,275
So with slides,
I guess this would kind of spawn
226
00:14:03,275 –> 00:14:06,478
some innovation as well
because you start to think, okay,
227
00:14:06,545 –> 00:14:09,181
why am I putting all this stuff
on individual slides?
228
00:14:09,181 –> 00:14:15,387
Where we’re going ultimately is lots of
smart people are working on technology
229
00:14:15,387 –> 00:14:18,824
so that the glass slides go away
eventually all together
230
00:14:18,824 –> 00:14:22,561
and that we’re actually able
to use the technology.
231
00:14:22,561 –> 00:14:25,331
And there’s several different ideas
that people are pursuing.
232
00:14:25,331 –> 00:14:29,101
But one idea is with something
called a micro CAT scan.
233
00:14:29,335 –> 00:14:33,072
So today we have big CAT scans
or CT scanners that you need
234
00:14:33,072 –> 00:14:36,775
to have a certain part of your body
scanned for better imaging.
235
00:14:36,909 –> 00:14:40,379
We’re all familiar with that
or my family members have one of those.
236
00:14:40,813 –> 00:14:45,217
We have small CT scanners
now that can sit on a countertop
237
00:14:45,751 –> 00:14:49,121
and that tissue sample itself
without the slide
238
00:14:49,121 –> 00:14:53,058
can be scanned by these micro CT scanners.
239
00:14:53,325 –> 00:14:57,363
And so from the original acquisition
240
00:14:57,696 –> 00:15:01,200
of the data, it’s in a digital format
and this is very similar
241
00:15:01,200 –> 00:15:06,105
to the early days of radiology,
where we would shoot a chest X-ray
242
00:15:06,238 –> 00:15:11,343
and then we would still have film
and we would put that film in a cartridge
243
00:15:11,677 –> 00:15:13,746
and then we would load that cartridge
244
00:15:14,113 –> 00:15:18,317
into a digitizer
and then it would digitally be scanned.
245
00:15:18,317 –> 00:15:21,420
And out comes this digital chest X-ray.
246
00:15:21,420 –> 00:15:25,724
So I liken the cartridge to the glass
slide today.
247
00:15:25,958 –> 00:15:31,363
And now when you have a chest X-ray, you
simply go into a digital radiology room.
248
00:15:31,597 –> 00:15:33,632
You still stand in front of a sensor.
249
00:15:33,632 –> 00:15:36,101
They shoot that X-ray image of you,
250
00:15:36,869 –> 00:15:39,538
but it’s completely digital in format.
251
00:15:39,538 –> 00:15:41,473
Now, from the point of acquisition.
252
00:15:41,473 –> 00:15:45,744
So from the point of acquisition
for pathology samples, in the future
253
00:15:45,878 –> 00:15:49,882
they will be digitized
versus having to have this interim
254
00:15:49,882 –> 00:15:53,585
medium of glass
for the tissue to be adhered to.
255
00:15:53,585 –> 00:15:55,654
So that’s where we’re ultimately going.
256
00:15:55,888 –> 00:15:57,756
I figured we wouldn’t
just keep doing the same thing
257
00:15:57,756 –> 00:16:00,859
over and over again because that does seem
very time consuming, very wasteful,
258
00:16:00,859 –> 00:16:03,796
and there are probably better ways
of doing this sort of thing
259
00:16:03,829 –> 00:16:05,097
that people are going to come up with.
260
00:16:05,097 –> 00:16:07,833
I’m not someone who’s going to come up
with that idea, obviously, but,
261
00:16:09,001 –> 00:16:09,268
you know,
262
00:16:09,268 –> 00:16:11,870
with my with my wipe off the slide idea
right
263
00:16:13,072 –> 00:16:16,442
away, no Clorox wipes, it’s fine.
264
00:16:17,843 –> 00:16:20,312
It’s like,
why can’t we repurpose that type of thing?
265
00:16:20,312 –> 00:16:23,315
And it is we don’t understand process
so why wouldn’t yeah.
266
00:16:23,382 –> 00:16:23,849
I mean it’s.
267
00:16:23,849 –> 00:16:27,186
I do think it’s really important
you know that evolution
268
00:16:27,486 –> 00:16:30,522
that will get us to
this digital age is really important.
269
00:16:30,522 –> 00:16:34,626
It is some super cool technology
to see what individuals and organizations
270
00:16:34,626 –> 00:16:36,595
are working on to move this field forward.
271
00:16:36,595 –> 00:16:41,533
When you say things like petabytes of data
and, you know, needing a place
272
00:16:41,533 –> 00:16:44,536
to put all that, I immediately think,
okay, well, where does NetApp fit in?
273
00:16:44,536 –> 00:16:44,670
Right.
274
00:16:44,670 –> 00:16:50,109
I mean, this sounds like a very lucrative
way for NetApp to get involved because
275
00:16:50,109 –> 00:16:53,612
we, of course, have multiple petabytes
to put things right.
276
00:16:53,912 –> 00:16:55,881
So where does not fit into all this?
277
00:16:55,881 –> 00:16:57,716
Are we looking at on-prem storage?
278
00:16:57,716 –> 00:16:59,251
Are we looking at cloud storage?
279
00:16:59,251 –> 00:16:59,885
A little of both?
280
00:16:59,885 –> 00:17:02,388
How is the industry moving towards this?
281
00:17:02,588 –> 00:17:06,358
Before I answer that question,
let me put this in perspective of what
282
00:17:06,358 –> 00:17:12,364
the impact to health care is from a health
IT perspective and the data management
283
00:17:12,364 –> 00:17:15,868
that’s going to be required
for this whole slide imaging.
284
00:17:15,868 –> 00:17:18,904
So we talked about the chest X-ray
a second ago
285
00:17:18,904 –> 00:17:21,907
and it’s a really great example
to compare.
286
00:17:21,907 –> 00:17:26,378
So a chest X-ray, normally
you get one or two views of that image.
287
00:17:26,378 –> 00:17:30,182
It’s very straightforward, single image
or maybe two images.
288
00:17:30,416 –> 00:17:34,353
Unlike an MRI or CT
where you put up hundreds of images
289
00:17:34,353 –> 00:17:37,022
that make up a single radiology exam.
290
00:17:37,456 –> 00:17:40,225
So a chest X-ray, one single image.
291
00:17:40,325 –> 00:17:43,929
It’s the highest resolution
image that we have in radiology.
292
00:17:44,296 –> 00:17:47,433
And it is approximately 10 to 15 megabytes
293
00:17:47,433 –> 00:17:50,736
for one radiology chest X-ray image.
294
00:17:51,003 –> 00:17:56,442
When we look at digital pathology
and whole slide imaging, a single image –
295
00:17:56,442 –> 00:18:02,147
and there will be multiple images
within a single case or exam –
296
00:18:02,514 –> 00:18:06,752
one full slide image is approximately one
297
00:18:06,752 –> 00:18:10,756
and a half to three gigabytes in size.
298
00:18:10,956 –> 00:18:15,461
So we can see that comparison
of the highest resolution, densest
299
00:18:15,461 –> 00:18:19,798
pixel image in radiology
today is 10 to 15 megabytes.
300
00:18:19,798 –> 00:18:25,304
We’lll jump to one and a half gigabytes
for a single pathology image.
301
00:18:25,304 –> 00:18:30,542
So CIOs know that they’re organized
and they’re starting to think about
302
00:18:30,843 –> 00:18:35,747
how do they begin to take this digital
transformation journey in pathology?
303
00:18:36,215 –> 00:18:38,717
And from a CIO perspective,
304
00:18:38,917 –> 00:18:41,987
how are they going to manage
all of this massive data?
305
00:18:41,987 –> 00:18:45,090
Because today it’s widely accepted that
306
00:18:45,090 –> 00:18:48,961
medical imaging comprises 80 – eight-zero
307
00:18:48,961 –> 00:18:52,931
percent – of all clinical data
generated worldwide.
308
00:18:53,332 –> 00:18:58,103
Well, that number, with the addition
of digital pathology, is set to explode.
309
00:18:58,737 –> 00:19:03,509
So if a medium sized community hospital
that’s generating
310
00:19:04,009 –> 00:19:08,680
100,000 pathology cases
a year is just average, right?
311
00:19:08,881 –> 00:19:14,253
That organization is looking at
probably 500 terabytes of data
312
00:19:14,253 –> 00:19:17,256
that’s going to be generated annually,
and that’s net-new data.
313
00:19:17,256 –> 00:19:21,960
Helping organizations understand
how to prepare for this
314
00:19:22,327 –> 00:19:25,230
new massive volume of data
that’s coming their way,
315
00:19:25,230 –> 00:19:28,467
as well as be able
to think about strategies
316
00:19:28,467 –> 00:19:33,572
to leverage the value of that data
in the use of AI model development
317
00:19:33,572 –> 00:19:39,077
or in the use of research and education
is a really great place for NetApp to be.
318
00:19:39,411 –> 00:19:43,949
And whether an organization
is looking to manage its data
319
00:19:44,283 –> 00:19:49,021
completely on-premises,
because we do still have some hesitations
320
00:19:49,021 –> 00:19:52,291
of putting masses of data in the cloud
from a health care perspective,
321
00:19:52,558 –> 00:19:55,727
whether that sort of a privacy
and security perspective
322
00:19:55,961 –> 00:19:57,296
or maybe there’s concerns
323
00:19:57,296 –> 00:20:00,832
about the egress fees if we do need
to bring that data down from the cloud.
324
00:20:01,133 –> 00:20:04,403
So whatever the reason,
if an organization wants to stay on-prem
325
00:20:04,736 –> 00:20:09,308
we certainly have great technologies
to help manage that data,
326
00:20:09,942 –> 00:20:14,546
whether it’s in that file format
as its initial type of data retention,
327
00:20:14,546 –> 00:20:18,884
or you want to promote that data
to a more cost effective object storage,
328
00:20:19,184 –> 00:20:20,552
perhaps using storage grade
329
00:20:20,552 –> 00:20:24,756
and using that fabric pools
to manage the tiering on-prem
330
00:20:25,123 –> 00:20:29,361
or if you are more cloud comfortable
or technologies can help
331
00:20:29,361 –> 00:20:34,800
you have that best hybrid experience
by being able to leverage our NetApp
332
00:20:34,866 –> 00:20:39,871
ONTAP operating system across
any of the public cloud providers.
333
00:20:39,871 –> 00:20:44,076
And it’s going to be really advantageous
and beneficial
334
00:20:44,476 –> 00:20:48,480
to the patient population
for organizations to be able
335
00:20:48,480 –> 00:20:55,454
to have their clinical data as part of the
AI algorithm development process
336
00:20:55,520 –> 00:20:59,524
or that AI model training or machine
learning model development.
337
00:20:59,691 –> 00:21:03,462
So organizations
that are wanting to leverage
338
00:21:03,462 –> 00:21:08,267
the value of their data
in that way, in a totally anonymized way
339
00:21:08,267 –> 00:21:11,370
that protects patient identity
and patient privacy.
340
00:21:11,370 –> 00:21:15,707
Of course, by leveraging NetApp
technologies, we can help organizations
341
00:21:15,707 –> 00:21:19,578
to replicate that information to the cloud
in a secure way.
342
00:21:19,578 –> 00:21:22,381
We’ve talked
about how NetApp fits in here and
343
00:21:22,981 –> 00:21:26,151
what sort of things digital pathology
brings to the industry.
344
00:21:26,785 –> 00:21:29,688
What about if I want to learn
more about this, like, where would I go
345
00:21:29,688 –> 00:21:32,891
to find more information
about digital pathology?
346
00:21:33,191 –> 00:21:34,393
NetApp Solutions?
347
00:21:34,393 –> 00:21:36,428
Yeah,
we have a great amount of information
348
00:21:36,428 –> 00:21:37,329
out there, Justin,
349
00:21:37,329 –> 00:21:38,563
that we’ll include the link
350
00:21:38,563 –> 00:21:42,401
in this podcast and then you can always
find this information at netapp.com.
351
00:21:42,601 –> 00:21:45,570
One thing that I did want to add
and one thing
352
00:21:45,570 –> 00:21:49,341
that we have information about that
I think is exciting and really timely for
353
00:21:49,341 –> 00:21:53,612
this field is a partnership
with Google Cloud that was announced
354
00:21:53,612 –> 00:21:57,115
at the beginning of October,
and this is the Google Cloud Medical
355
00:21:57,115 –> 00:22:00,185
Imaging suite of which NetApp
is a core component of.
356
00:22:00,185 –> 00:22:04,790
And with the Google Cloud Medical
Imaging Suite, we’re able to replicate
357
00:22:05,123 –> 00:22:09,061
data volumes from on-premises,
whether that’s an entire dataset
358
00:22:09,361 –> 00:22:12,831
or perhaps a data cohort,
replicate that data
359
00:22:12,831 –> 00:22:16,501
into the Google Cloud,
into a cloud volumes, ONTAP
360
00:22:17,102 –> 00:22:21,707
instance or Cloud Volumes Service instance
and leveraging technologies
361
00:22:21,707 –> 00:22:25,243
like Cloud Data Sense,
we’re able to help customers
362
00:22:25,510 –> 00:22:29,147
very quickly
be able to categorize and map the data
363
00:22:29,147 –> 00:22:34,453
that’s contained within the medical image
or this in this case, a whole slide image
364
00:22:34,886 –> 00:22:38,557
so that they can easily be able
to audit the data.
365
00:22:38,724 –> 00:22:43,662
But also in this use case, most
importantly, be able to query the data
366
00:22:44,129 –> 00:22:47,866
and create specific AI data cohorts
367
00:22:48,133 –> 00:22:51,503
that then feed into this greater
368
00:22:51,770 –> 00:22:54,873
AI model development environment
within the Google Cloud
369
00:22:55,073 –> 00:22:58,910
that provides a robust set of tools
to help customers
370
00:22:58,910 –> 00:23:03,648
be able to partner with third party
AI developers, in secure ways,
371
00:23:03,882 –> 00:23:08,353
so that those developers do not have to
actually be accessing the data on site.
372
00:23:08,820 –> 00:23:12,457
So by leveraging this
Google Medical Imaging suite,
373
00:23:12,457 –> 00:23:17,129
we are able to bring the power of NetApp
technologies, partner with Google,
374
00:23:17,129 –> 00:23:22,968
to provide a rich set of AI development
tools and services for customers
375
00:23:22,968 –> 00:23:25,904
to be able to take their pathology data
to the next level.
376
00:23:26,104 –> 00:23:28,940
Sounds like we got a good start
on this burgeoning industry, right?
377
00:23:29,441 –> 00:23:33,078
It’s not quite at the level of a
medical imaging, but it is getting there.
378
00:23:33,078 –> 00:23:35,313
Right. It’s the future of pathology there.
379
00:23:35,313 –> 00:23:38,750
So, Kim, again, if we wanted to reach you,
how do we do that?
380
00:23:38,884 –> 00:23:42,954
Yeah, you can simply reach me
at kim.garriott@netapp.com.
381
00:23:43,188 –> 00:23:45,857
All right.
That music tells me it’s time to go.
382
00:23:45,857 –> 00:23:47,492
If you’d like to get in touch with us,
send us an email
383
00:23:47,492 –> 00:23:51,296
to podcast@netapp.com,
or send us a tweet @NetApp.
384
00:23:51,730 –> 00:23:55,500
As always, if you’d like to subscribe
find us on iTunes, Spotify,
385
00:23:55,567 –> 00:24:00,605
Google Play, iHeartRadio, SoundCloud,
Stitcher or via techontappodcast.com.
386
00:24:01,072 –> 00:24:03,108
If you liked the show
today, leave us a review.
387
00:24:03,108 –> 00:24:06,711
On behalf of the entire Tech ONTAP podcast
team, I’d like to thank Kim Garriott
388
00:24:06,812 –> 00:24:07,813
for joining us today.
389
00:24:07,813 –> 00:24:14,386
As always, thanks for listening.
390
00:24:27,365 –> 00:24:30,802
[Tech
391
00:24:30,802 –> 00:24:34,239
ONTAP
392
00:24:34,239 –> 00:24:37,676
Podcast
393
00:24:37,676 –> 00:24:41,112
Outro
394
00:24:41,112 –> 00:24:44,549
theme]
Why Is The Internet Broken? 
Pingback: A Year in Review: 2022 Highlights | Why Is The Internet Broken?