Big data and AI in a radiologist's reading room. Check out this presentation by the EuSoMII president Sergey Morozov.(MD, PhD, MPH, CEO of Radiology Research and Practical Centre, Chief Radiology Officer of Moscow, President of Moscow RSR).

1. www.eusomii.pro
Professor Sergey Morozov
Eketerina Guseva
Big data and AI
in a radiologist’s
reading room

2. www.eusomii.pro
Agenda
• Radiology trends
• How to measure imaging quality
• Moscow URIS
• How to manage changes
• AI/ML for imaging data analysis

3. www.eusomii.pro
3
morozov@npcmr.ru

4. www.eusomii.pro4

5. www.eusomii.pro
Radiology trends
• From modalities (CT, MRI, X-Ray, etc.) to
clinical sections
• From analog to digital
• From a dark room to open space
• From jack-of-all-trades to labor division
• From separated to integrated clinical facilities
• From art to craft and skills
• From unique services to mass production

6. www.eusomii.pro
Structure + process + team =
= value for the patient
Technician
Radiologist
Administrator
Head of department
Consultant
Nurse
Referring
physician

7. www.eusomii.pro
Familiar problems?
• Shortage of qualified staff
• Low motivation
• “Cherry picking” principle
• High staff turnover
• Slow implementation of new methods
France, Imaging management, 2009

8. www.eusomii.pro
Quality problem
• IoM, USA: 100 000 deaths/year of medical
errors
8
Original report
According to CT, the tumor of the middle pole of the
left kidney. Cyst of the left kidney. Multiple myomas
in the uterus.
Peer-review
Cyst of the left kidney. Multiple myomas in the
uterus.
False positive CT diagnosis of renal cancer

9. www.eusomii.pro
Radiology Team =
“Defensive Radiology”

10. www.eusomii.pro
Quality Domains
§ Appropriateness
§ Diagnostic accuracy
§ Safety
§ Actionable reporting
§ Communication
Measuring Radiologist’s
Value Add

11. www.eusomii.pro

12. www.eusomii.pro
77 med centers
62 CT
40 MRI140
radiologists
+ 22 experts
RPC Medical Radiology
26 mammo
United RIS (Radiology Information Service)
3 PET/CT
Administrative department:
+ ORACLE dashboard
Teaching and consulting dpt:
+ Teleaudit
+ Moodle CME
15 X-Ray5 fluorograph
> 700 000 studies

13. www.eusomii.pro
Teleaudit of CT, MRI, mammo
Reports
Ratings
Bonuses
Webinars
CME
Random 5%
selection of
studies
Anonymized
remote
viewing
Expert’s
detection of
errors

14. www.eusomii.pro
14
Competition among
CMOs of hospitals
Positive motivation
to achieve the
effectiveness goals
Managing results by ratings
ГБУЗ "ГП № 64 ДЗМ"
ГБУЗ "ГП № 66 ДЗМ"
ГБУЗ "ГП № 69 ДЗМ"
ГБУЗ "КДЦ № 2 ДЗМ"
ГБУЗ "ГП № 195 ДЗМ"
ГБУЗ "ГП № 209 ДЗМ"
ГБУЗ "ГП № 212 ДЗМ"
ГБУЗ "ГП № 8 ДЗМ"
ГБУЗ "КДЦ № 4 ДЗМ"
ГБУЗ "ГП № 45 ДЗМ"
ГБУЗ "ГП № 6 ДЗМ"

15. www.eusomii.pro
Target audience
• Radiologists
• Clinicians
• Technicians
03-2016 05-2017
0
50
100
150
200
250
300
350
Jan – Apr 2017
• 45 webinars
• 6885 attendees from 20
regions of Russia
Webinars of RPC Medical Radiology
I.Trofimenko
i.trofimenko@rpcmr.org.ru
Free webinars of RPCMR

16. www.eusomii.pro
9,4
22,6 20,5
47,5
4,6
14
10,4
69
0
10
20
30
40
50
60
70
80
Клинически
значимые
расхождения
Клинически не
значимые
расхождения
Замечания
общего
характера
Полное
соответствие
I кв. 2016 I кв. 2017
Out-patient medical centers
Teleaudit results
16
Clinically
significant
Not clinically
significant
General
remarks and
recommx
No remarks

17. www.eusomii.pro
What’s next?

18. www.eusomii.pro
• “Does your work feed into the
field of artificial intelligence?”
• “Not really, we study natural
stupidity instead of AI”
Amos Tversky

19. www.eusomii.pro
Healthcare automatization
• Total technical potential ≈ 36%
• Nutrition of patients, distribution of
medicines> 75%
• Collecting information from patients> 50%
• Analysis of pictures ≈ 30%
• Nurse <30%
• Dental hygienist ≈ 13%
19morozov@npcmr.ru

20. www.eusomii.pro
X-Ray/Lab
USD/lab
PET/lab
MRI-CT/lab
Radiologist
Labor automation potential
http://www.visualcapitalist.com/charting-automation-potential-of-u-s-jobs/

21. www.eusomii.pro
#AIHealth
DIAGNOSTICS AND
RISK ANALYTICS DRAW
MOST DEALS IN 2016
Imaging & diagnostics
dominated 2015, and is
at par with insights &
risk analytics so far this
year.
Deals to lifestyle
management &
monitoring picked up in
2015.
Virtual assistants, and AI
startups in mental health
still have room for
further investment
activity.
www.cbinsights.com 16
AI IN HEALTHCARE: CATEGORY HEATMAP
2012-2016 (as of 8/30/2016)
No. of deals
Min Max
CATEGORY 2012 2013 2014 2015 2016YTD
Insights & Risk Analytics
Lifestyle Management & Monitoring
Imaging & Diagnostics
Drug Discovery
Emergengy Room & Hospital Management
Wearables
Virtual Assistants
Mental Health
Research
Nutrition
Miscellaneous
www.cbinsights.com
Investments in AI/ML for
imaging
21
morozov@npcmr.ru
Enlitiс.com
develops AI/ML
• X-ray/CT/MR
• CXR AUC 0.92
• Radiologist’s
AUC 0.5-0.72
Automatic lung nodes detection,
measurement, classification, and
reporting

22. www.eusomii.pro
Self-driving MRI?
Self-driving Radiology
department?

23. www.eusomii.pro
AI/ML business
model – yet to
be defined

24. www.eusomii.pro
Infrastructure
readiness is
required

25. www.eusomii.pro
Healthcare preparedness for change?

26. www.eusomii.pro
26
● Up to 36% of medical functions in healthcare can be
automated
● Healthcare automation brings cost savings of appr.
7-11%
● Artificial Intelligence algorithms’ development starts
from a problem, but not a technical solution
● Most of developed AI algorithms are useless
● Infrastructure and data
● Expert-as-a service is a business model of the future
● Key competency of the future – change
management
Take-home points

27. www.eusomii.pro

28. www.eusomii.pro