[March 21, 2018] |
|
Intel Editorial: One Simple Truth about Artificial Intelligence in Healthcare: It's Already Here
The following is an opinion editorial provided by Navin Shenoy,
executive vice president and general manager of the Data Center Group at
Intel (News - Alert) Corporation.
This press release features multimedia. View the full release here:
https://www.businesswire.com/news/home/20180321006119/en/
Navin Shenoy, executive vice president and general manager of the Data Center Group at Intel Corporation, speaks Wednesday, March 21, 2018, at the University of California, San Francisco. Intel invited leading healthcare organizations to address the most pressing topics and challenges in healthcare. (Credit: Intel Corporation)
In the wide world of big data, artificial intelligence (AI) holds
transformational promise. Everything from manufacturing to
transportation to retail to education will be improved through its
application. But nowhere is that potential more profound than in healthcare,
where every one of us has a stake.
What if we could predict the next big disease epidemic, and stop it
before it kills? What if we could look at zettabytes of data to find
those at greatest risk of becoming sick, then quickly and precisely
prevent that from happening? What if the treatment and management of
chronic disease could be so personalized that no two individuals get the
same medicine, but equally enjoy the best possible outcome? What if we
could drastically reduce the time and cost to discover new drugs and
bring them on the market? What if we could do all of that now?
Thanks to artificial
intelligence and the work of Intel
and its partners, we can.
Real Impact Today
There's a common myth that AI in healthcare is the stuff of science
fiction - think machines diagnosing illness and prescribing treatment
without a doctor involved. But that is not only highly unlikely, it's
not even close to the best examples of how AI is emerging in healthcare
today.
Intel and partners throughout the healthcare industry - including GE
Healthcare, Siemens,
Sharp
Healthcare, the Broad
Institute, UCSF
and the Mayo
Clinic - are successfully applying AI solutions today, from the back
office to the doctor's office, from the emergency room to the living
room. A few customers that we're working closely with include:
Montefiore
Medical System: using prescriptive models to identify
patients at risk for respiratory failure, so healthcare workers can act
on alerts that lead to timely interventions that save lives and
resources.
Stanford Medical: using AI to augment MRI image reconstruction so
that a complete image can be delivered in about a minute versus what
normally would take about an hour - eliminating risky intubation and
sedation in pediatric patients during imaging exams.
ICON plc: Instead of only relying on burdensome clinic visits and
paper diaries, using clinical data from sensors and wearable devices to
more quickly assess the impact of new therapies in clinical trials.
AccuHealth: Using home monitoring along with data mining and
predictive modeling to identify changes of concern among chronic disease
patients to enable intervention before conditions escalate and become
acute.
Better Health Tomorrow
But the triumph of artificial intelligence in healthcare isn't
inevitable. Right now, the average hospital generates 665 terabytes of
data annually,1 but most of that data isn't useful. At least
80 percent of hospital data is unstructured,2 such as
clinical notes, video and images. Electronic medical records (EMRs) are
a mandated system of record, but they aren't as actionable as they could
be. Only with AI can we leverage healthcare data to create a system of
insights.
Getting healthcare systems to provide greater access to their data would
help. Government also has a role to play by providing apropriate
incentives and regulatory clarity for sharing data. We agree with the
recent White
House proposal to give patients control and ownership of all their
health data, bringing it with them wherever they go rather than residing
in various doctor's offices, clinics and hospitals.
New technology can help as well. One example: Intel researchers are
making great strides toward practical methods for homomorphic
encryption, a method that will allow computer systems to perform
calculations on encrypted information without decrypting it first. Such
encryption would enable researchers to operate on data in a secure and
private way, while still delivering insightful results.
Indeed, much work is ahead, and Intel is uniquely
positioned to help healthcare organizations succeed. Emerging
healthcare data is massive data - images, a growing list of 'omics (i.e.
genomics, proteomics), video - and will require a storage plan and a
network that addresses speed, latency and reliability. We have been
investing with our partners to build the right systems - data, storage,
network, full infrastructure - all the way from the edge to the network
to the cloud, and everywhere in between. With the advancements in our
hardware and optimizations of popular deep learning frameworks, the
Intel Xeon Scalable processor has 198x better inference performance and
127x better training performance than prior generations3. As
a result, the Xeon platform is at the center or many AI workloads that
are real today because it is well suited for many machine and deep
learning applications across industries like healthcare.
But hardware, storage and network alone are not enough. We need to
leverage the unparalleled expertise from data scientists, software
developers, industry experts, and ecosystem partners --to address AI in
healthcare end to end. As part of the effort to expand expertise across
AI, we launched the Intel AI Academy, a place that offers learning
materials, community tools and technology to boost AI developments. With
more than 250K monthly participants, I invite you to join
for free as well.
I feel very fortunate to work for a company like Intel that is committed
to powering AI solutions that will tackle some of the biggest challenges
of our time, including healthcare. I'm also proud to be leading the team
that will deliver that vision.
____________________
1 Source: http://www.netapp.com/us/media/wp-7169.pdf
2 Source: http://www.zdnet.com/news/unstructured-data-challenge-or-asset/6356681
3 Source: Configuration: AI Performance - Software + Hardware
(see below)
-
INFERENCE using FP32 Batch Size Caffe GoogleNet v1 256 AlexNet 256.
-
Performance estimates were obtained prior to implementation of recent
software patches and firmware updates intended to address exploits
referred to as "Spectre" and "Meltdown." Implementation of these
updates may make these results inapplicable to your device or system.
Software and workloads used in performance tests may have been
optimized for performance only on Intel microprocessors. Performance
tests, such as SYSmark and MobileMark, are measured using specific
computer systems, components, software, operations and functions. Any
change to any of those factors may cause the results to vary. You
should consult other information and performance tests to assist you
in fully evaluating your contemplated purchases, including the
performance of that product when combined with other products. For
more complete information visit http://www.intel.com/performance
Source: Intel measured as of June 2017 Optimization Notice: Intel's
compilers may or may not optimize to the same degree for non-Intel
microprocessors for optimizations that are not unique to Intel
microprocessors. These optimizations include SSE2, SSE3, and SSSE3
instruction sets and other optimizations. Intel does not guarantee the
availability, functionality, or effectiveness of any optimization on
microprocessors not manufactured by Intel. Microprocessor-dependent
optimizations in this product are intended for use with Intel
microprocessors. Certain optimizations not specific to Intel
microarchitecture are reserved for Intel microprocessors. Please refer
to the applicable product User and Reference Guides for more
information regarding the specific instruction sets covered by this
notice.
-
Configurations for Inference throughput
-
Processor :2 socket Intel(R) Xeon(R) Platinum 8180 CPU @ 2.50GHz / 28
cores HT ON (News - Alert) , Turbo ON Total Memory 376.46GB (12slots / 32 GB / 2666
MHz).CentOS Linux-7.3.1611-Core , SSD sda RS3WC080 HDD 744.1GB,sdb
RS3WC080 HDD 1.5TB,sdc RS3WC080 HDD 5.5TB , Deep Learning Framework
caffe version: f6d01efbe93f70726ea3796a4b89c612365a6341 Topology
:googlenet_v1 BIOS:SE5C620.86B.00.01.0004.071220170215 MKLDNN:
version: ae00102be506ed0fe2099c6557df2aa88ad57ec1 NoDataLayer.
Measured: 1190 imgs/sec vs Platform: 2S Intel® Xeon® CPU E5-2699 v3 @
2.30GHz (18 cores), HT enabled, turbo disabled, scaling governor set
to "performance" via intel_pstate driver, 256GB DDR4-2133 ECC RAM (News - Alert).
CentOS Linux release 7.3.1611 (Core), Linux kernel
3.10.0-514.el7.x86_64. OS drive: Seagate* Enterprise ST2000NX0253 2 TB
2.5" Internal Hard Drive. Performance measured with: Environment
variables: KMP_AFFINITY='granularity=fine, compact,1,0',
OMP_NUM_THREADS=36, CPU Freq set with cpupower frequency-set -d 2.3G
-u 2.3G -g performance. Deep Learning Frameworks: Intel Caffe: (http://github.com/intel/caffe/),
revision b0ef3236528a2c7d2988f249d347d5fdae831236. Inference measured
with "caffe time --forward_only" command, training measured with
"caffe time" command. For "ConvNet" topologies, dummy dataset was
used. For other topologies, data was stored on local storage and
cached in memory before training. Topology specs from https://github.com/intel/caffe/tree/master/models/intel_optimized_models
(GoogLeNet, AlexNet, and ResNet-50), https://github.com/intel/caffe/tree/master/models/default_vgg_19
(VGG-19), and https://github.com/soumith/convnet-benchmarks/tree/master/caffe/imagenet_winners
(ConvNet benchmarks; files were updated to use newer Caffe prototxt
format but are functionally equivalent). GCC 4.8.5, MKLML version
2017.0.2.20170110. BVLC-Caffe: https://github.com/BVLC/caffe,
Inference & Training measured with "caffe time" command. For "ConvNet"
topologies, dummy dataset was used. For other topologies, data was st
ored on local storage and cached in memory before training BVLC Caffe (http://github.com/BVLC/caffe),
revision 91b09280f5233cafc62954c98ce8bc4c204e7475 (commit date
5/14/2017). BLAS: atlas ver. 3.10.1.
-
Configuration for training throughput:
-
Processor (News - Alert) :2 socket Intel(R) Xeon(R) Platinum 8180 CPU @ 2.50GHz / 28
cores HT ON , Turbo ON Total Memory 376.28GB (12slots / 32 GB / 2666
MHz).CentOS Linux-7.3.1611-Core , SSD sda RS3WC080 HDD 744.1GB,sdb
RS3WC080 HDD 1.5TB,sdc RS3WC080 HDD 5.5TB , Deep Learning Framework
caffe version: f6d01efbe93f70726ea3796a4b89c612365a6341 Topology
:alexnet BIOS:SE5C620.86B.00.01.0009.101920170742 MKLDNN:
version: ae00102be506ed0fe2099c6557df2aa88ad57ec1 NoDataLayer.
Measured: 1023 imgs/sec vs Platform: 2S Intel® Xeon® CPU E5-2699 v3 @
2.30GHz (18 cores), HT enabled, turbo disabled, scaling governor set
to "performance" via intel_pstate driver, 256GB DDR4-2133 ECC RAM.
CentOS Linux release 7.3.1611 (Core), Linux kernel
3.10.0-514.el7.x86_64. OS drive: Seagate (News - Alert)* Enterprise ST2000NX0253 2 TB
2.5" Internal Hard Drive. Performance measured with: Environment
variables: KMP_AFFINITY='granularity=fine, compact,1,0',
OMP_NUM_THREADS=36, CPU Freq set with cpupower frequency-set -d 2.3G
-u 2.3G -g performance. Deep Learning Frameworks: Intel Caffe: (http://github.com/intel/caffe/),
revision b0ef3236528a2c7d2988f249d347d5fdae831236. Inference measured
with "caffe time --forward_only" command, training measured with
"caffe time" command. For "ConvNet" topologies, dummy dataset was
used. For other topologies, data was stored on local storage and
cached in memory before training. Topology specs from https://github.com/intel/caffe/tree/master/models/intel_optimized_models
(GoogLeNet, AlexNet, and ResNet-50), https://github.com/intel/caffe/tree/master/models/default_vgg_19
(VGG-19), and https://github.com/soumith/convnet-benchmarks/tree/master/caffe/imagenet_winners
(ConvNet benchmarks; files were updated to use newer Caffe prototxt
format but are functionally equivalent). GCC 4.8.5, MKLML version
2017.0.2.20170110. BVLC-Caffe: https://github.com/BVLC/caffe,
Inference & Training measured with "caffe time" command. For "ConvNet"
topologies, dummy dataset was used. For other topologies, data was st
ored on local storage and cached in memory before training BVLC Caffe (http://github.com/BVLC/caffe),
revision 91b09280f5233cafc62954c98ce8bc4c204e7475 (commit date
5/14/2017). BLAS: atlas ver. 3.10.1.)
*Other names and brands may be claimed as the property of others.
View source version on businesswire.com: https://www.businesswire.com/news/home/20180321006119/en/
[ Back To Mobile World Congress's Homepage ]
|