Presentation that gives an overview of the impact of IT on radiology, including the growing role of biomarkers and artificial intelligence and deep learning on the (future) radiology profession. The shift to precision medicine and personalized care are explained, the reasons for a re-definition of radiology are addressed.
2. Evolution
• Major innovations in IT and healthcare in past decades
1. Internet & broadband connections
2. Digitisation of medical imaging
3. Rise of mobile networking and data exchange
• Major impact on radiology
– The filmless, full digital department is now reality
– Images go beyond the radiology department
– Images are not property of radiologists anymore
– Any-ology is developing
– Teleradiology applications are growing
– Standards & guidelines for sharing images & patient data efficiently, safely and securely
3. Any-ology
• Era of image sharing
• Other specialties have needs
for both static & cine image
storage & retrieval
• Added value for collaboration
• More unified management of
data and images needed
• Yet not sufficiently addressed
by current methodologies
Radiology Dermatology Ophthalmology
Oncology Cardiology Pathology
Otorhinolaryngology Neurosurgery Gastro-enterology
3Source: SIIM Webinar Jan 28, 2016: “Real world challenges and benefits of cloud technologies”
4. New IT-developments in medicine
and radiology
• e-Health
• IoT
• Cloud-based services
• Personalised care
• Precision radiology
• Imaging biomarkers
• Image-guided therapy
• Artificial Intelligence, deep learning
5. e-Health
• Internet has become an indispensable gateway for electronic
transmission & sharing of health-related data, a process
known as “e-Health”
• The electronic health record (EHR) is being introduced, which
allows to keep a longitudinal and complete electronic record
of the patient’s health information
6. IoT
• A growing number of electronic devices
and sensors is becoming connected to the
Internet and gradually shaping the
Internet ofThings (IoT).
• Those wireless “smart wearables” can be
used for a wide variety of health-related
purposes, such as monitoring of heart
rate, body temperature, mobility, sugar
levels, etc.
• The IoT enables real-time collection of an
enormous amount of autonomous health-
related data, even on a global scale
7. Personalised care
• The volume of electronically shared health data is increasing
exponentially
• The power for processing these data as well.
• By cross-linking information obtained from intelligent cloud-
based data analysis with enormous amounts of genetic data,
it has become possible to generate information that is useful
for providing personalised care
8. Personalised Disease Evaluation
• Although evidence based medicine has succeeded in defining effective
therapeutics for large populations, it is lacking when applied to small
subpopulations (“precision medicine”) and ultimately to the individual
level (“personalized medicine”).
• Evidence-based medicine is “old fashioned” now
• What do we want to know for each patient?
– Is a tumor present?
– Is it aggressive?
– Is it focally treatable?
– Is it radiosensitive?
– Will it metastasize?
Krishnaraj A et al.The future of imaging biomarkers in
radiologic practice.
J Am Coll Radiol 2014;11:20-23
10. Precision radiology
• In radiology the importance of Imaging biomarkers is increasing.
• Radiologists must incorporate new biological developments and research into
clinical reality through use of images and their data
• Radiology can participate in precision medicine by
– Depicting early abnormalities
– Predicting prognosis
– Grading abnormalities
– Defining FU outcomes
• These elements facilitate clinical decision-making for personalised care
https://www.evernote.com/shard/s91/sh/1c5712eb-1e95-4551-be3b-451d3f940f13/ed9b74075b3fc1db670a9b609b1e81da
11. Imaging biomarkers
• "VIRTUAL biopsies” : non-invasive evaluation in 3D of tissues over time.
– Structural (e.g. measurements of volume changes and shape).
– Metabolic (e.g. maps of rate of glucose metabolism or distribution of hypoxic
cells in a tumor).
– Physiological (e.g. maps of regional blood flow and vessel permeability).
– Composition (e.g. measuring the regional distributions of choline).
– Molecular (e.g. mapping the density of specific cell receptors in a tumor).
– Cellular (e.g. evaluating tumor cell density or viability).
– Biophysical (e.g. assessing variations in tissue elasticity).
13. Radiomics & radiogenomics
• It will be possible to make therapeutic decisions based upon the
combination of morphologic information from medical images
with genomic data, a technology known as “radiogenomics”.
• The term “radiomics” is referring to the automated morphologic
analysis of radiological images with new cloud-based deep-
learning techniques, converting these images to mineable data.
• The term “radiogenomics” is preferably used for the process of
correlating the data obtained from radiomics with genomic
(genetic) information of a disease and/or patient.
14. “To explore the full potential of
radiomics, we have to enter the era
of big data, team science and, most
of all, the new age of imaging
bioinformatics”
Dr. Hricak
15.
16.
17. Image-guided treatment
techniques
• New image-guided treatment techniques are being developed,
e.g. image-guided radiation therapy (IGRT).
• Repeated imaging is performed during the treatment to identify
changes in the tumour’s size and location, allowing adjustment of
the patient’s position and/or the radiation dose.
• This can increase the accuracy of radiation treatment and may
allow reductions in the planned volume of tissue to be treated,
thereby decreasing the total radiation dose.
18. MRI-guided radiotherapy
• At theVUmc Cancer Center Amsterdam, the Netherlands, the first
radiation therapy machine in Europe with integrated MRI scanner
was installed.
• Advantage: implantation of gold particles for marking the tumour
becomes unnecessary, since with MRI it iss possible to discern
healthy soft tissue more accurately from tumour than with plain
CT.
• This indicates that a more active engagement of the radiologist
will be required, also in the treatment process.
20. New HC model
• The increasing user-based demand for access to digital
data is causing a gradual degradation of the traditional
health care model, in which the hospital plays a central
role.
• The walls surrounding the traditional hospital-based
information-silos are progressively being “deconstructed”,
which causes a shift from the classic hospital-centric model
towards a more patient-centric model of care.
21. Shift of HC model
Hospital-centric model
Personalised
care &
genomics
Computer-
aided
diagnosis &
treatment
IoT
Patient data
consolidation
and
integration
Open
Patient
involvemen
t
Patient-centric model
Hospital centric
Episodic
Departments
Proprietary
Data silos
Patients
22. Liquid Hospital
• In this model all relevant patient data are
shared fluently between all stakeholders of
the healthcare process.
• The information stream needs to become
more “liquid”, so that it can run as easy as
water in a river.
23. Artificial Intelligence
• AI made a huge step forward in recent years.
• IBM is developing the highly intelligent software with code
name “Avicenna”, which is based on the IBM-Watson
computer system
• The company plans to leverage theWatson Health Cloud
“to analyse and cross-reference medical images against a
deep trove of lab results, electronic health records, genomic
tests, clinical studies and other health-related data sources”
25. What is deep learning?
• Inspired by how brain works
• Higher layers form higher levels of
abstraction
26. What is deep learning?
• Deep Learning and autoencoders
are mimicking human brain
activity and can automatically
identify patterns in a dataset.
• In the Google Brain project
autoencoders successfully
trained themselves to recognise
human and cat faces based on 10
million digital images taken from
YouTube videosMulti-layered neural networks
27. Deep learning in radiology
Image processing
• Deep learning algorithms will
help select and extract
features from medical images
as well as construct new ones;
• This will lead to
representations of imaging
studies as never seen before.
Image interpretation
• Deep learning will help not
only to identify, classify, and
quantify disease patterns
from images,
• but will also allow to measure
predictive targets and create
actionable prediction models
of care pathways.
http://ecronline.myesr.org/ecr2016/index.php?p=recorddetail&rid=9d74262a8c7e5843d02c85391c774654&t=browsesession
s)
29. Artificial
Intelligence
• Deep Learning chest X-rays
classifier
• Developed over GPU
technology (Tesla K40 from
NVIDIA)
• 4200 annotated chest X-rays
• Sensitivity: 83%
• Specificity: 80%
• …and improving with more
cases
30. How will fast AI go?
• Many deep-learning algorithms still need to be
developed, tested and approved before it will be
possible to implement AI routinely for clinical
purposes.
• It can be expected that in 5 to 10 years from now
Avicenna or a similar AI system will be sufficiently
trained to act as a first filter for analysing all sorts of
medical images that are later examined by doctors.
32. Ownership
• It is should be questioned to what extent in the future
image analysis will be performed by computers
instead of radiologists, and what effect this will have
on the “ownership” of the technology.
• Such evolution could possibly translate into a
challenge about the value of the work and the
financial compensation of radiologists.
33. Replacement of radiologists?
• Supercomputers could act as a provider of second
opinions, helping to confirm a radiologist’s suspicion of an
unusual or difficult diagnosis.
• This in turn could cut down on tedious work, superfluous
and unnecessary testing,
• Hereby saving time for the patient, eliminating
unnecessary radiological exposure, reducing radiologists’
workloads and reducing costs.
34. Solution
• Radiologists should start with embracing AI as soon as they can, with the
main intention to participate in AI-research, with the objective of creating
IT-tools that add can value to radiology services.
• Radiologists should try to use AI for managing their workload more
efficiently.
• AI could be used to do preliminary reads of imaging studies for example,
so that radiologists are able to use Watson’s information to make their
final report
• Or do we want to build skyscrapers with 100-year old tool such as “image
reporting”?
35. AI becomes IA
• By doing so radiological error rates could possibly be
reduced.
• AI should be regarded at as a form of intelligence
amplification (IA)
– A technique enabling radiologists to add value to the
radiology report.
– A technique to consolidate the radiologists’ role, instead
of replacing them.
36. Future PACS
• The future PACS will be a "portal"
to radiological knowledge:
– incorporation of quantitative
imaging methods
– Integration of patient data
– automatic retrieval of images
similar images to those under
review
– Decision support and artificial
intelligence to support radiologistSource: Laboratory of Imaging Informatics, Stanford University
http://www.stanford.edu/~rubin/projects.html
37. Communication
• Enabling patients to contact the radiologist for explaining
the findings could create a greater awareness of the crucial
role of radiologists.
• With iPortals (e.g. POW) providing access to radiology
reports this will become a requirement.
• It could be considered to use “multimedia reports”, in
which the information is displayed in a simplified but more
structured and interactive manner
40. What business do we have?
• Value-creation business
– More than traditional “reading”
– “Move from mindset of reading images to creating and
organising information for greater accuracy, faster speed
and lower cost in medical decision-making” (Giles Boland, MGH)
• Business model should focus on collaboration, patient-
centric approach and value-based services
Boland GW, Duszak R, McGinty G, Allen B. Delivery of appropriateness, quality, safety, efficiency and patient satisfaction. JACR. 2014;11(1):7-11.
41.
42. Future role of radiologists
• "Radiologists won't simply be interpretors of imaging studies, they
will be the curators of quantitative and descriptive data about
disease processes that will enable computerized decision-support
systems to improve diagnostic and prognostic accuracy".
• "Radiologists will identify the volumes and areas of interest that can
be segmented. From these volumes, computers can extract
hundreds of descriptive quantitative features. These features can
then be combined with medical and genomic data to create a
comprehensive database."
RSNA News, Feb 2015
43. Redefine radiology
• Easy & fast sharing of data
• Embracement ofAI
• Imaging for personalised care
• Image-guided treatments
• Patient communication
– e-Consultation
– Multimedia reports
Eliot Siegel, schooling IBM’s Watson at the University of Maryland;
Image courtesy RSNA.org
44. Adapt
• Radiologists will want to be more than just adopters of new algorithms
• They will have to become creators & designers of new algorithms
• Radiologists with engineering skills will have strong influence on
successful integration of machine-learning into radiology workflow
• Several workflow-issues must be optimised to create new value, e.g.
– To determine indications for using algorithms
– How to interpret and integrate the algorithm’s outputs
– How to monitor the utilisation of algorithms
45. Summary
• The radiologist as manager should make use of
information technology (IT) in 4 main domains:
1. the management of workflow,
2. the interpretation of images,
3. the treatment of patients and
4. the communication with clinicians and patients
46. Patient
Patient Data, EHR
Dynamic
workflow
monitoring and
management
Image analysis
Artificial Intelligence
Disease monitoring
Communication
social media,
tele-consultation,
online scheduling
Treatment
decisions,
Image-guided
therapy
Radiologist
Manager radiology services
Workflow Interpretation Treatment Communication
47. Conclusion
• The development of machine-learning applications in radiology is
building momentum with intense speculation about what this means for
the profession, much is still unproven.
• We must not ignore, we must but embrace it, to ensure the best outcome
for radiologists and patients.
• Radiology stands at an exhilarating crossroad, and we have an incredible
opportunity to transform this specialty into more than we ever imagined
it could be.
Other “ologies” are following fast – entering digital era of imaging, needs better more unified ways to manage the data
Cloud-based enterprise storage for all –ologies
Challenge in silo –model!
Cloud has single point of access, promise of using virtualisation!
Gebruiken van mobile devices:
Communicatie met collegae en patiënten
Professionele apps
Educatieve doeleinden, informatie bij de hand
De radioloog wordt manager van radiologische services op 4 verschillende vlakken
Workflow – complexe procedures moeten vereenvoudigd worden; SIMPLIFICATIE
Beoordeling – Gebruik van “omics” data (genomics, proteomics, radiomics), integratie van HOLISTISCH patiëntbeeld van patiënt obv EHR/EPD = CONTEXT IS KING, progressieve omschakeling van AI naar IA = INTELLIGENCE AMPLIFICATION
Behandeling - begeleiden van “verpersoonlijkte” zorgprocessen, minder-invasieve beeldgeleide behandelingen,
Communicatie – omschakeling naar ECHTE patient-centred services door meer directe communicatie tussen radioloog en patiënt, software benutten om meer tijd vrij te maken voor de patiënt
De beschikbaarheid van en toegang tot ALLE patiëntgegevens is hierbij van cruciaal belang en slechts 1 onderdeel van het geheel;
Daarnaast is ook verregaande integratie van data noodzakelijk – INTEROPERABILITY of UITWISSELBAARHEID is de sleutel voor de toekomst
Patient beslist uiteindelijk wie toegang heeft/krijgt tot deze informatie en wanneer. De toegang tot patiëntinformatie is mobiel inzetbaar, onafhankelijk van tijd en locatie.