In this talk, Dr. Fangde Liu will discuss how deep learning technology is used in surgical robots originally for neurosurgery and later adapted to orthopaedics.
He'll share his view on the opportunities and issues of applying deep learning and autonomous technology for surgical application.
3. NEXT CHAPETER
OF AI
BIG IDEA
GREAT POTENTIAL
CONVINCING THE
PUBLIC
GAME CHANING APPLICATION
1) DRIVERLESS CAR
2) FINACE
3) MEDICINE
4) EDUCATION
5) AGRICUTRUE
4. MEDICINE AND
HEALTHCARE
1)BIG
NCD cost 30 Trillion dollars,
HEALTHCARE 25% of US GDP
2) SHORTAGE
70% is Labour cost
1/3 doctor shortage in the
US
3) Evidence /Experience
Based
Economic Burden
Mortality
7. SURGICAL AUTONOMY (PROS AND
CONS)
2) Controlled Environment1) Narrow AI
3) Human for Help
Accuracy And Risk
Skill and Art (not
science)
Time is Critical
Valuable
4) Cost Insensitive
8. SURGICAL AUTONOMY
(NOT VERY DIFFERENT FROM
DRIVERLESS CAR)
TECHNICALLY ACHIEVABLE
Semantic
segmentati
on
Pose
estimatio
n
Motion
plannin
g
Robot/US
9. SURGICAL AUTONOMY
(PERCEPTION)
Medical Imaging Processing
(Tracking, Segmentation,
Registration)
DL makes Medical Images Processing as good as Human
Very Good Results even before DL (statistic shape model)
11. THE DARK HISTORY OF SURGICAL
AUTONOMY
Full Autonomous, 1997 No Autonomy, 2000
cooperative, 2001
Technology Is Available long
before
Surgeon Hate Them All
1) No Partial Solution
2) No Complete Solution
Supporting Role
Do Nothing
12. REASON FOR AI FAILURE
MEDICINE PROBLEM (NO
DISCRIMINATIVE BIOMARKER, NOT
EFFECTIVE TREATMENT) VS
ENGINEERING PROBLEM
EFFICIENCY? VS EFFICIENCY
VEHICLE VS DRIVER
Machine Learning can be only as good as
Human.
Your Problem is usually Surgeons don’t do
well.
INFORMATION MISSING VS EFFICIENCY
13. MY RECIPE (FOR NON-
GENIUS)
1) Calculating Not Machine Learning
(Beat Human in Accuracy and Speed)
Surgeons are bad at maths
2) Take Vehicle Advantage
Find a better surgical procedure
Impossible for human
3) Machine Learning for Simple job.
Not for magic.
Do many Simple Job.
Reason for Machine Learning is for the autonomy more complete,
4) Don’t Do
Anything Intra-
operatively
thinking three
times
1) Never Fail
2) No Risk
3) Useful
Don’t Do cutting!5) Check the
medicine
14. WHAT CAN WE DO(PATIENT SPECIFIC
INSTRUMENTATION)(AI For Autonomous Tool Design)
15. Patient Specific Rehearsal (AI For VR Content
Generation)
Clinician Need VR for Rehearsal
(senior surgeons really want)
Most VR Provides Training
(senior surgeons don’t care)
Content
Generation
17. CONCLUSI
ON
Medicine is Hard,
AI has Great Potential, AI can do many things for Medicin
Still Deliver Clinical Impact is Very Hard.
Have the Big Picture is Very Important.
For Medical and also for Other Industry
Autonomy:
people are looking for something smart enough for any c
But stupid enough to take whatever he orders