SURGICAL:AI AI Democratise Computer Aided Surgery WHY NOT GOOGLE - - PowerPoint PPT Presentation

SURGICAL:AI

AI Democratise Computer Aided Surgery

WHY NOT GOOGLE MAP AMAZON UBER FACEBOOK ALPHAGO FOR SURGERY

SURGERY CRISIS

Healthcare Cost $30 Trillions (US 23% GDP) Surgery Cost 40% ($12 Trillion) 5 Billion have no Access to Surgery

OA Affects 1 in 5 (US), 1 in 6 (China) OA cost 1%-2.5% GDP, Second cause of disability Surgery serves Only 5% patients (US), 0.5% (China)

SURGICAL AUTONOMY

AI and Robotics 50% , worth $6 Trillion Narrow AI Highly Controlled Environment Human Support Cost Insensitive Ethical , Risk and Regulation

DO SURGEON REALLY NEEDS AUTONOMY

ProBot Full Autonomous, 1997 Da Vinci, No Autonomy, 2000 ACRobot/Mako/Stryker Cooperative, 2012 BrainLab, Navigation Sedasys Robotic Drill

THE BARRIERS (T,C,S,A)

Time (prepare) Time (OR) Capabilit y Safety Affordabi lity Manual 10 minutes 2 hour 100 experts 50% success +0 Robot +1 Day 4 hour 1,000 experts 90% success +$500,00- $2,00,000 Navigation +1 Day 3 hour 300 experts 75% success +$100,000

• 300,00

PSI +1 Month 1 hour 1,000 experts 90% success +$1,000 SurgicalAI 10 minutes 1 hour 1000 experts 95% success +$500

SURGICAL:AI

AUTONOMOUS PSI DESIGN BY AI

Design in Less in 10 seconds, Deliver in 4 hours Web base Mixture Reality Interface Deep Morphology Engine Planning And Design Engine TensorLab: Big Data Warehouse And Autonomous Machine Learning System

SHOULDER PSI EXAMPLE (CAOS 2016) (FULLY AUTONOMOUS)

Deep Morphology Engine 1 Data Scarcity: 30 Volumes vs 2 Millions 2 Memory Consumption 3D volume vs 1 D signal 3 No Explanation and No Confidence 4 Flexible Workflow….. 5 Not Cooperative Traditional Deep Learning for Medical Images processing Wishing A baby will becoming a great doctor by showing him lots of scans without going to school

= + +

Neural Poincare Engine Deep Poincare Map (https://arxiv.org/pdf/1703.09200.pdf) It works like Surgeon, It has memory, knows where to attend its focus Speed: 0.2 Seconds for Segmentation one bone from a volume Data Efficiently: trained 5 volume we achieve accuracy close to human Memory Efficient: Needs about 500MB memory, It can run on your mobile phone Flexible: Parts scan or arbitrary posture. Error Estimation: We know when it fails.

SURGICAL DB

Medical Image and Processed Images PSI Models Implants The Surgery Plan

NoSQL and Map Reduce Search Engine Solution

Find me all the pre and after surgery CT scans of all the patients using Company A implants on the third Spine Joint

The only Question Surgeons Care but No existing PACS or EHR system can Answer

TensorLab for Autonomous Continuous Learning

Heart Failure Acc IoU Time Recruitment by Search Engine The Principal: 1) Everything is Data. 2) Everything is Identified by Query. 3) Pulling based Stream Processing Images and Sematic Labels Model Versions and Performance Metrics Left Volume Model Architecture Search Engine is used as the Router for Data Job1 Job 2 Job3 Job4 Job5 Job6 Builder Analyser Validator Builder Analyser Builder Analyser Validator Job Queue Job Query as Load Balancer Build Job

Data: Heart with Tumour

Arch: Best Segmenation Model

Input Version: Most Accurate Version Output Version: Heart Tumour Detection

CLINICAL RESULTS:

Save 40 minutes in OR,1000GBP for 1 procedure

PATIENT SPECIFIC OSTEOTOMY

Guide was designed, printed and sterilized Guide was fixed to the unique position on the bone Bone was resected following the guidance of the guide 15° wedge was resected and removed Wedge was closed and bone was fixed

OTHER PROCEDURES

APPLICATIONS

1) Surgery Planning 2) Patient Specific Guide Device 4) Implants Registration and Management 3) Surgical Workflow Management

A TEAM OF WORLD LEADING EXPERTS

Professor Anan Shetty (Chief Medical Scientist)

Dr Fangde Liu (CEO &CRO)

The best Surgeon in the UK Shetty-Kim Technique Inventor The Surgeon of 2017 by Royal College of Surgeon Proudest Award for UK, the Hunterian Professor Advisor of NICE ( National Institute of Clinical Excellence) Chief Architecture of Several Flagship Surgical Robots, Pharmacovigilance for multination Autonomous CMR Analysis System and Diagnosis Senior Associate of Royal Society of Medicine Dr Hao Dong (Engineering VP) PhD at Imperial College, Data Science Institute Deep Learning Engineering Expert The Creator of TensorLayer Most Popular DL Framework in China Yinna Bu (Client and BD Manager in China) MSc from UCL CEO of two start-ups

Prof Yike Guo (Advisor) Prof Ferdinando Rodriguez y Baena (Advisor) Dr Joshua Giles (Advisor) Dr Jianmo Li (Advisor) Full Professor of Medical Robots at IC, World First Orthopaedics Surgical Robots Sold For Stryker 1.67 Billion Dollars Team Leader of Biggest Surgical Robots Project In European Union Chair of World Computer Aided Orthopaedics Full Professor of Data Science at Imperial College Chairman of TranSmart Foundation World Standard Platform for Translation and Clinical Medicine Adopted by all the major Pham Company in the world Including J&J, Merkek, Roche, Sanophie. PostDoc at Imperial College Inventor of Should PSI PhD Prof Andrew Amis Postdoc at Imperial College Inventor of Knee PSI Mr Saif Ahmed Consultant Surgeon Professor of Anatomy Expert on 3D Printing and Surgical Robots Executive Comminute of Computer Aided Orthopaedics' UK

Kent Institute of Medicine & Surgery

• KIMS

is the largest Independent hospital in SouthEast England and Kent.

• No.1

patient satisfaction (92%) against other 92 independent hospital in the UK

• The hospital is equipped with state-
• f-the-art technology and innovative

diagnostic services

• Wide range of treatment – covering

all the orthopedic area and joint replacement

AI Democratise Computer Aided Surgery Autonomous Surgery is Feasible The barrier is Human Machine Cooperation For Surgery, Not Just AI, Vehicle Matters Provides the tool, not solution GPU plays crucial by breaking the time barrier

CAOS 2016, Japan Osaka IDEALondon, 2016 Daily Mail, 2016 Re.Work Deep Learning in Healthcare Summit, 2017 GTC 2017

Surgical AI: Bring Big Data and AI into Operating Theatre Thank You