Anatomic and Computational Pathology Diagnostic Artificial Intelligence at Scale

Size: px

Start display at page:

Download "Anatomic and Computational Pathology Diagnostic Artificial Intelligence at Scale"

Catherine White
6 years ago
Views:

General Hospital Partners Healthcare System Harvard Medical School MGH

1 Anatomic and Computational Pathology Diagnostic Artificial Intelligence at Scale John Gilbertson MD Department of Pathology Massachusetts General Hospital Partners Healthcare System Harvard Medical School MGH - Partners Fellowship in Clinical Informatics 5/25/2017 API Gilbertson 1

2 Notice of Faculty Disclosure In accordance with ACCME guidelines, any individual in a position to influence and/or control the content of this ASCP CME activity has disclosed all relevant financial relationships within the past 12 months with commercial interests that provide products and/or services related to the content of this CME activity. The individual below has disclosed the following financial relationship(s) with commercial interest(s): John Gilbertson, MD Philips Strategic Advisory Boards fees & expenses Sunquest SAB/EAB expenses Inspirata Strategic Advisory Board fees and stock options 5/25/2017 API Gilbertson 2

3 Anatomic and Computational Pathology Diagnostic Artificial Intelligence at Scale John Gilbertson MD Department of Pathology Massachusetts General Hospital Partners Healthcare System Harvard Medical School MGH - Partners Fellowship in Clinical Informatics 5/25/2017 API Gilbertson 3

4 Today s Talk Developing Deep Learning Applications in each Anatomic Pathology Subspecialty at Partners Scale Deep Learning Projects AP Sub-specialties Instigators Engagement A Broad Base of Machine Learning Expertise How do we get an Infrastructure that Supports Scale Two Themes Why do we care about Engagement 5/25/2017 API Gilbertson 4

5 Why Involvement and Engagement Informatics & Molecular are the areas of Innovation Mickey Mouse Model circa 2007 Molecular Informatics Clinical Laboratories They should infuse the Pathology services with new ideas and techniques That would transform the way we practice medicine and understand disease Research Informatics as a transformational agent 5/25/2017 API Gilbertson 5

pathologist think of disease Molecular concepts were increasingly defining Pathology s vision of itself Research

6 Molecular was being embraced by a large number of pathologists in all subspecialties Pathologists understood Molecular was and how it can be used Molecular Clinical Laboratories Informatics becoming central to the way pathologist think of disease Molecular concepts were increasingly defining Pathology s vision of itself Research Anatomic and Molecular Pathology (formerly AP) Laboratory and Molecular Medicine (formerly CP) 5/25/2017 API Gilbertson 6

But Informatics Informatics is not part of the intellectual fabric of Pathology There is little consideration of its potential value There is little engagement Molecular

7 But Informatics Informatics is not part of the intellectual fabric of Pathology There is little consideration of its potential value There is little engagement Molecular Clinical Laboratories Research Informatics Anatomic and Computational Pathology (formerly AP) Laboratory and Computational Medicine (formerly CP) 5/25/2017 API Gilbertson 7

8 Computational Pathology An Emerging Definition A statement of potential value Pathology as Meteorology A Call to Change Practice -- Engagement of pathologist is vital Archives of Pathology /25/2017 API Gilbertson 8

9 Computational Pathology Computational Pathology in Laboratory Medicine Most Implementations are in Clinical Pathology LAB LIS Analytical Integration Visualization Computational Engines EMR WARD Computation is taking off in Clinical Pathology despite challenges 5/25/2017 API Gilbertson 9

10 Computational Pathology Computational Pathology in Anatomic Pathology The Problem and the Opportunity Signout Room Histology Lab Analytical Integration Visualization Computational Engines 80% of our Pathologists are in Anatomic Pathology 5/25/2017 API Gilbertson 10

WSI was seen an enabling technology for Computational Pathology before there was Computational

and at high fidelity, we can apply computational power and network connectivity, the driving

study of tissue and the practice of Anatomic Pathology WSI s long term value is in the

11 WSI was seen an enabling technology for Computational Pathology before there was Computational Pathology Ul Balis John Sinard If we can digitize all of our slides, easily, rapidly and at high fidelity, we can apply computational power and network connectivity, the driving forces of innovation, discovery, and productivity in the modern world, to the morphologic study of tissue and the practice of Anatomic Pathology WSI s long term value is in the enabling of diagnostic image analysis APIII circa Interscope, /25/2017 API Gilbertson 11

12 Computational Pathology WSI as the Enabler of Computational Anatomic Pathology Pathologists Control the Entire Process Histology Lab WSI Analytical Integration Visualization Computational Engines Signout Room At this only works if the Scanners work and all the Pathologists are on board 5/25/2017 API Gilbertson 12

Anatomic and Computational Pathology The Importance of the FDA Studies Over the several years we (and others) have worked with Philips on studies that cumulated with the FDA de novo authorization for

13 Anatomic and Computational Pathology The Importance of the FDA Studies Over the several years we (and others) have worked with Philips on studies that cumulated with the FDA de novo authorization for primary diagnosis Four major studies (and several small ones) Many MGH pathologist were involved Experienced very low re-scan rates, high precision, high reliability (and non-inferiority) we developed confidence in the devices and the processes Which meant we could begin to consider WSI as a engine of change in Pathology 5/25/2017 API Gilbertson 13

Anatomic and Computational Pathology So what did we consider doing We did not consider digitizing our clinical workflow We focused on the develop of computational tools to enhance pathology s

14 Anatomic and Computational Pathology So what did we consider doing We did not consider digitizing our clinical workflow We focused on the develop of computational tools to enhance pathology s capabilities and its Vision of Itself Three converging areas of interest: Imaging as a Diagnostic and Development Platform Images as Data Structures Diagnostic Deep Learning: is very pathology like, allows for multiple projects 5/25/2017 API Gilbertson 14

15 Developing Deep Learning Applications in each Anatomic Pathology Subspecialty at Partners Scale Deep Learning Projects AP Sub-specialties Instigators Engagement A Broad Base of Machine Learning Expertise 5/25/2017 API Gilbertson 15

Anatomic and Computational Pathology Engagement of the Pathologists Machine Learning in all pathology subspecialities and all clinical labs Informatics

16 Anatomic and Computational Pathology Engagement of the Pathologists Machine Learning in all pathology subspecialities and all clinical labs Informatics Diaspora Teams (pathologists, informaticians, data scientists, technicians, datasets) from each subspecialty) Circa /25/2017 API Gilbertson 16

17 Today s Talk Developing Deep Learning Applications in each Anatomic Pathology Subspecialty at Partners Scale Deep Learning Projects AP Sub-specialties Instigators Engagement A Broad Base of Machine Learning Expertise How do we get an Infrastructure that Supports Scale Two Themes Why do we care about Engagement 5/25/2017 API Gilbertson 17

18 Anatomic and Computational Pathology Strategic Considerations, Technical Partners, and Pathology Department Resources Strategic Considerations: Unified Pathology Vision & Enterprise Interest Technical Partners: Strong, Long Term Industrial and Academic Partnerships Pathology Resources: Specimens, Data, Domain Experts 5/25/2017 API Gilbertson 18

19 Strategic Conditions (1 of 2) Unified Pathology Vision Anatomic and Computational Pathology Strategic Conditions, Technical Partners and Pathology Department Resources LIS Architecture drives social change Six Departments of Pathology ecare ecare ecare C C D S Initial State Partners Enterprise Pathology Imaging AI 5/25/2017 API Gilbertson 19

Strategic Conditions (2 of 2) Anatomic and Computational Pathology Strategic Conditions, Technical Partners and Pathology Department Resources Enterprise Vision for

20 Strategic Conditions (2 of 2) Anatomic and Computational Pathology Strategic Conditions, Technical Partners and Pathology Department Resources Enterprise Vision for Informatics and AI Diagnostics: - The World Medical Innovation Forum : Neurology and Psych : Oncology : Cardiovascular :????? 5/25/2017 API Gilbertson 20

21 Technical Partners (1 of 3) Anatomic and Computational Pathology Strategic Conditions, Technical Partners and Pathology Department Resources Strong, Internal, Industrial Partnerships: Strong, engaged industrial partners are fundamental to computational and imaging success Strong Industrial Partnerships make your academic Informatics better: Write Paper Yes Identify a Need Build a System Implement Locally Works? No Implement Somewhere Else? Works? 5/25/2017 API Gilbertson 21

22 Technical Partners (2 of 3) Anatomic and Computational Pathology Strategic Conditions, Technical Partners and Pathology Department Resources Strong Internal Academic Partnerships: Partners Center for Clinical Data Sciences: - AI based applications in Medicine - Created in Pathology is on the Board Does your institution have something like this? - If your AMC has one, call it - If your AMC doesn t, somebody wants one - Image & Data Management - Helps manage policy - Compute (1500 Teraflops) - Recruitment of AI Talent - Funding - Connections with Industry 5/25/2017 API Gilbertson 22

Partners and Pathology Department Resources East Cambridge

23 BOSTON MAP Local Technical Partners (3 of 3) Anatomic and Computational Pathology Strategic Conditions, Technical Partners and Pathology Department Resources East Cambridge MGH East Cambridge MIT External Partners 5/25/2017 API Gilbertson 23

24 Pathology Department Resources: Anatomic and Computational Pathology Strategic Conditions, Technical Partners and Pathology Department Resources Pathology Department Resources: In Machine Learning, data drives everything In Machine Learning, data drives everything Anatomic Pathology has the specimens, slides, data and domain expertise If you can provide this, you can drive everything With a scanner and policy, you can be a significant player 5/25/2017 API Gilbertson 24

25 Deep Learning Collaborative Groups Center for Clinical Data Sciences IMS --- Image Management Platform --- Compute Scanners -- Tracking --- Specimens ---- LIS Subspecialty Pathology Groups Partners Enterprise Pathology - MGH and BWH Existing Principal Investigator Datasets 5/25/2017 API Gilbertson 25

26 Anatomic and Computational Pathology Fellowship and Training If you are a resident in Anatomic Pathology today... Don t change you disciple, change the way you think of it AP will be digital, computational and innovative Academic and Business Models will change... Knowledge will be in algorithms, software and data A Blended Fellowship in Informatics and Anatomic Pathology is well designed for an anatomic pathologist of the future Relative Value in a Project Technology Domain Expertise Technology Maturation 5/25/2017 API Gilbertson 26

27 Anatomic and Computational Pathology Fellowship and Training If you are a resident in Anatomic Pathology today... Don t change you disciple, change the way you think of it AP will be digital, computational and innovative Academic and Business Models will change... Knowledge will be in algorithms, software and data 1 cm A Blended Fellowship in Informatics and Anatomic Pathology is well designed for an anatomic pathologist of the future 5/25/2017 API Gilbertson 27

28 Conclusions - Key Points The importance of large scale involvement in Anatomic Computational Pathology The importance of the FDA studies (and DICOM) The architecture of the initiative The major components if this initiative are available in many AMC today It is a great time to be in Pathology Informatics 5/25/2017 API Gilbertson 28

Color aspects and Color Standardization in Digital Microscopy

Color aspects and Color Standardization in Digital Microscopy Yukako Yagi, PhD yyagi@partners.org Director of the MGH Pathology Imaging & Communication Technology Center Assistant Professor of Pathology,