1

Towards a Pathology Imaging Data Commons for Discovery and Precision Oncology

AACR SY32: Integrative Data Science for the Precision Medicine EraApril 17th 2018

Michael J. Becich, MD PhD, FACMIChair, Department of Biomedical InformaticsUniversity of Pittsburgh School of Medicine

University Distinguished Professor,Associate Vice Chancellor for Informatics

Associate Director for Cancer Informatics, Hillman Cancer CenterNCI Board of Scientific Advisors

2

Becich Conflicts of Interest (& Disclaimer)• Spatial Diagnostics, Inc. or SPDx (founder and stock) –

computational pathology newco• Nexi – Newco by Rebecca Jacobson and Text Information

Extraction System (TIES) Cancer Research Network (TCRN) team (licensing revenues to my Department)

• Cancer Center Consultancies and EABs – Baylor, Cancer Institute of New Jersey, Indiana University, Northwestern Lurie Cancer Center, University of Colorado, UCLA, University of Michigan, University of New Mexico and Wake Forest

• CTSA Consultancies and EABs – numerous (not a conflict for this talk except possibly for U of Chicago Institute for Translational Medicine)

• Funding – CDC/NIOSH, NCATS, NCI, NHGRI, NHLBI & NLMDisclaimer – I am a member of NCI’s Board of Scientific Advisors and Frederick National Laboratories Advisory Committee, Technical Working Group

3

Outline• Introduction to Text Information System (TIES)

Cancer Research Network (TCRN) • Introduction to Whole Slide Imaging (WSI) and

Computational Pathology (Comp Path)• Introduction to Pathology Imaging Data Commons

Enabled by TCRNKey Impact Areas – Comp Path Predictive Analytics for Precision Oncology leads to better health and

discovery science!

http://cancerdatanetwork.org/

4

The Text Information Extraction System or TIESNCI ITCR funded effort (U24 CA180921)

• A natural language processing (NLP) and Information Retrieval system for de-identifying, annotating, storing & retrieving pathology (and radiology) reports

• A system for indexing research resources (clinical data, biospecimens & images) with document annotations

• An GUI for querying large repository of annotated documents and obtaining resources locally, using an honest broker model

• A platform to support phenotype, images and biospecimen sharing among networks of cancer centers and other institutions

http://ties.dbmi.pitt.edu/

5

TIES Cancer Research Network (TCRN)

http://cancerdatanetwork.org/

6

TIES Cancer Research Network (TCRN)UPMC Hillman Cancer Center (lead) • Augusta University Cancer Center • Abramson Cancer Center (Penn)• Roswell Park Cancer Institute• Stonybrook University (new partner)• Sidney Kimmel Cancer Center (TJU)• 12 Cancer Centers interested in joining

Network Trust Agreements• IRBs agree that use of data for

investigators is Not Human Subjects Research, no need for an additional IRB protocol even to access record level de-id data

• Governance• Agreement to abide by SOPs• Instrument of Adherence

Soliciting new WSI “ready” partners!

http://cancerdatanetwork.org/

7

TIES Cancer Research Network

http://cancerdatanetwork.org/

8

8

9

TCRN Impact Merkel Cell Cancer – Rare Cancer Use Case

http://cancerdatanetwork.org/

10

Adding Cancer Registry Data to TIESOutcomes Annotation

• Identified as a valued development target by users• We have secured additional funding from Institute

for Precision Medicine in Pittsburgh• Mike Davis and Jonathan Silverstein lead this effort• Starting with breast cancer (first publication in

review)• Immediately leveragable by existing TCRN nodes by

adding Cancer Registry data to TIES instances

http://cancerdatanetwork.org/

11

Cancer Registry Data ElementsDemographics Primary Treatment Outcome

Race Primary Site Surgery Vital Status

Gender Histology Chemotherapy Cancer Status

Age @ Diagnosis

Grade BRM Recurrence

Smoking Path TNM Hormonal Cause of Death

Alcohol Clinical TNM Immunotherapy

Prognostic Factors (including site specific)

Rad Onc

http://cancerdatanetwork.org/

12

13

• Kaiser team publication – 2017 J. Path Info article:

• Result = deeper patient annotation, including outcomes data and enabling study cohort identification

Validation of TIES/TCRN Cancer Registry Integration

http://cancerdatanetwork.org/

14

Led by Rebecca Jacobson, MD MSIS

http://ties.dbmi.pitt.edu/tcrn/

15

Outline• Introduction to Text Information System (TIES)

Cancer Research Network (TCRN) • Introduction to Whole Slide Imaging (WSI) and

Computational Pathology (Comp Path)• Introduction to Pathology Imaging Data Commons

Enabled by TCRNKey Impact Areas – Comp Path Predictive Analytics for Precision Oncology leads to better health and

discovery science!

16

Traditional vs. Digital PathologyIntro to Whole Slide Imaging (WSI)

From Bertram and Kopfleish, Vet Path 2017

17

WSI – Imaging Big Data Drives Computational Pathology

From Saltz, circa 2011, to NLM

18

WSI Comp Path Adds Spatial Big Data for Tumor Analysis

EpigenomicsGenomics Proteomics Metabolomics

1 mm

Multi to hyperplexed fluorescence imaging of whole slide images for higher spatial resolution

and tissue context

H&E stained whole slide imagesfrom FFPE tumor sample

1

4

4 2A

B

C

D

23

13

Thanks to Drs. Joe Ayoob and Chakra Chennubhotla for this illustration, 2017

NCI ITCR funded effort (U01 CA204826)

19

Outline• Introduction to Text Information System (TIES)

Cancer Research Network (TCRN) • Introduction to Whole Slide Imaging (WSI) and

Computational Pathology (Comp Path)• Introduction to Pathology Imaging Data Commons

Enabled by TCRNKey Impact Areas – Comp Path Predictive Analytics for Precision Oncology leads to better health and

discovery science!

20

• Clinical Phenotype – From Pathology Lab Info Sys– Structured data– Unstructured data via NLP (Text Info Extract Sys – TIES)

• Outcomes Data – From Cancer Registry Systems• Biobank Data – From Formalin Fixed Paraffin

Embedded Tissue• Imaging – WSI - Spatial Annotation – HTAN relevant• Integrate Challenge – Clinical, Outcome, Biobank,

Imaging Data for analysis with Genomic data

Towards Computational Pathology Imaging Commons for Cancer – TCRN Data Types

http://cancerdatanetwork.org/

21Marusyk A. Nat Rev Cancer. 2012 Apr 19;12(5):323-34

Computationally Annotated WSI Assist inQuantification of Tumor Heterogeneity

22

Intra-tumoral spatial heterogeneity complicates accurate diagnosis & prognosis

Nature Reviews Cancer (2009) 9:239–252

23

Tumor Heterogeneity Research Interactive Visualization Environment (THRIVE – NCI ITCR U01 @ Pitt)

• Computational Pathology– Hyperplex Immunofluorescence (9 to > 50 Ab) – Machine Learning + Spatial Statistics– Network Systems Biology– Partnership with General Electric (GE)

• Iterative experimental–computational tumor micro-environment studies

Courtesy Chakra Chennubhotla, 2017

24

THRIVE – Chennubhotla Lab

Cancer Res. 2017 Nov 1;77(21):e71-e74. doi: 10.1158/0008-5472. PMID: 29092944

25

TCRN Enabled Data Sharing and Query

26

Comp Path Impact – TILs in TCGA Tumors

Cell Reports, 2018

NCI ITCR Funded U24 CA180924 (Saltz and Sharma), U24 CA215109 (Saltz and Prior) and others

27

Conclusions• The Text Information System (TIES) Cancer Research

Network (TCRN) is an ideal platform for coupling clinical/outcomes data to biospecimens/images

• Whole Slide Imaging (WSI) and Computational Pathology will enable NCI’s Moonshot in new ways to couple spatial analysis features to –omic data sets

• A WSI Imaging Data Sharing and Query Toolkit (TCRN) with linkages to NCI ITCR and GDC programs can assist the NCI vision for a Cancer Research Data Commons

Key Impact Areas – Computational Pathology Predictive Analytics for Precision Oncology leads to better health

and discovery science!For Copy of PPT –becich@pitt.edu

28

Computational Pathology @ PittLed by Chakra Chennubhotla, PhD

• Organized as a Interest Group and Lecture Series• Currently 109 members from Pitt, CMU, UPMC and

regional companies/startups (need sponsor!!!)• Pittsburgh Computational Pathology Interest Group

- http://www.csb.pitt.edu/comppath/• Computational Pathology Lecture Archive -

https://www.youtube.com/channel/UCWfBS3PLWHTIACeccm2CIog

• Supported by Akif Burak Tosun, PhD (post-doc)

29

Computational Pathology @ Pitt

Computational Pathology Lecture Archive -https://www.youtube.com/channel/UCWfBS3PLWHTIACeccm2CIog

30

“...Data commons collocate data, storage, and computing infrastructure with core services and commonly used tools and applications for managing, analyzing, and sharing data to create an interoperable resource for the research community...”

From Computing In Science & Engineering IEEE 2016

31

FAIR Data Principles

• Findable• Accessible• Interoperable• Reusable

• Data Use Case for Data Commons– See Data Med v3.0 – http://datamed.org

32

Six Requirements of a Data Commons

• Permanent digital IDs (data and knowledge)• Permanent metadata (data describing data)• API (interface)-based access (interoperability)• Data portability (standard containers)• Data Peering (commons 1 can access commons 2)• Pay for compute (allocate computing/charging)

– Demand higher than computing resources availableFrom Grossman et al 2016 CISE IEEE

33

Big Data to Knowledge (BD2K)

Pitt’s Department of Biomedical Informatics is a Center of Excellence in Big Data to Knowledge

Department of Biomedical Informatics

Pittsburgh, your Self-Driving Uber is arriving

now!!!

34

Informatics Fuels Data to Knowledge

35

HTAN and Spatial Imaging Discovery

36

Human Cell Atlas & Human BioMolecular Atlas Program (HuBMap) Drive Relevance of Comp Path

37

Enabling Research on the Cancer Registry

TIES tranSMART