THRIVE transmitted light services
TCRNSpatial
Characteristics of TIL
Distribution
THRIVEQuIPFeature DB
# of TILs inAll
200 Tumor Samples
QuIP analysis services
Detection of Tumor Infiltrating Lymphocytes
Spatial heterogeneity
features
Loca
l Pro
cess
Rem
ote
Acce
ss
The TIES Cancer Research Network (TCRN): Computational Pathology Support for Precision Oncology
Michael J. Becich, Jonathan Silverstein, Michael Davis, Joyce Zelnis, Degan Hao, Melissa Schwenk, Faina Linkov, Nita Maihle, Roni Bollag, Ashish Sharma, Carmelo Gaudioso, Chakra Chennubhotla, Akif Burak Tosun, Tahsin Kurc, Joel Saltz, Jack London, Mathias Brochhausen, Lawrence Tarbox, Fred Prior, Michael Feldman and John Gilbertson
TIES Cancer Research Network (Augusta U, Emory U, RPCI, Stonybrook, Thomas Jefferson U, UAMS, U Penn, U Pitt)
Figure 3 – TCRN ArchitectureTCRN Framework and Trust Fabric
University of PittsburghDepartment of Biomedical Informatics
Advances in cancer research and personalized medicine require significant new bridging infrastructures, including more robust biorepositories that link human tissue to clinical phenotypes and outcomes data. The Text Information Extraction System (TIES) (http://ties.dbmi.pitt.edu) Cancer Research Network (TCRN) (https://cancerdatanetwork.org) is a novel platform developed at the Department of Biomedical Informatics, University of Pittsburgh that enables cancer researchers to mine the content of pathology reports and to share biospecimens across a federated network. Member sites can access pathology data that are de-identified and processed with the TIES natural language processing system (NLP), which creates a repository of rich phenotype data linked to clinical biospecimens. TIES incorporates multiple security and privacy best practices that, combined with legal agreements, network policies, and procedures, enable regulatory compliance. The institutional trust fabric TCRN has developed has been “road tested” by our network and has been used by several funded translational clinical research programs.
Abstract
Each node in the infrastructure includes sharing of whole slide images (WSI) linked to pathology report data (NLP), tissue specimens and outcomes data (Cancer Registry). Quantitative Imaging in Pathology (QuIP) and Tumor Heterogeneity Research Interactive Visualization Environment (THRIVE) modules will provide imaging features (e.g., cytologic/spatial heterogeneity, host immune response) to enhance deeper search capability for cohorts and specimens. Software service tiers:
A deployed imaging analysis pipelines implemented in QuIP and THRIVE to generate rich sets of imaging features from whole slide images (WSI). Imaging features will be computed for image patches at the image level (i.e., WSI will be partitioned into patches, and imaging features and classification labels will be generated for each patch). The image-level features will be aggregated and summarized at the tumor level. QuIP is used to carry out image classifications and computation of features from segmentation of nuclear material in images. THRIVE is used to identify and quantify spatial ITH and infer interactions between different cell phenotypes and non-cellular constituents.
Innovation
TCRN has established an innovative inter-institutional trust fabric, based on TCRN security and regulatory compliance best practices and a TIES adoption blueprint that allows federation of data and biospecimenresources, but these network agreements must be modified to allow for HIPAA-compliant open data sharing with the NCI community in partnership with TCIA. An established Policy Working Group for TCRN composed of one member from each of the participating network sites who will establish a path to modifying and implementing this new network agreement and will further develop the existing Material Transfer and Data Use Agreements to allow researchers to efficiently access data and biospecimens. TCRN currently has nearly 7 million fully de-identified, annotated pathology reports linked to tissue specimens.
*
SUSTAINABLE OPEN SOURCE SOFTWARE ARCHITECTURE(via TIES refactored version 6.0 w/ microservices & APIs)
NETWORK SITES• UPMC Hillman
Cancer Ctr• Stony Brook Cancer Ctr• Abramson Cancer Center• Roswell Park Cancer Inst• Georgia Cancer Center• SKCC-TJU• To-Be-Named Site #1• To-Be-Named Site #2• Unfunded Additions
CURATED COLLECTONS• TCRN Studies
at Hub• Other Private
CollectionsDirectory
Indexes
Annotations
REMOTEPROCESSING
• *QuIP• *THRIVE• Other Open
Source Tools(e.g., Xena, PIIP)
IDENTIFIED DATA AT NETWORK SITES• Cancer Registry• Pathology Reports• Whole Slide Images
(WSI)
LOCAL PROCESS• TIES NLP• *QuIP• *THRIVE
Se
cu
re L
oca
l D
ep
loy
me
nt
TCRN Portal
Co
ntr
oll
ed
Acce
ss
Cu
rati
on
Da
tath
on
s&
Ha
ck
ath
on
s
Findable Accessible Interoperable Re-usable
DISCOVERY• Cohorts• Insights• Knowledge
Tumor Microenvironment
Immuno-Oncology
Tumor Heterogeneity
Other Areas of
Cancer Research
TCRN Portal
QU
ER
Y
OPEN COLLECTIONS• *TCIA• Other Public
CollectionsDirectory
Indexes
Annotations
Op
en
Sh
ari
ng
RE
ST
AP
I E
nd
po
ints
See Fig. 3
+Network &
oPotential
Network Sites
FIRE
WALL
Black: Aim 1Blue: Aim 2Violet: Aim 3Green: Aim 4; existing & expanding TCRN*Asterisks denote the 5 interacting ITCR U24 Programs
+
+
++ ++
FAIR
*TIES/TCRN POLICY FRAMEWORK AND TRUST FABRIC(from *Globus)
Private Data Stores
Research (“open”)
Data Stores
De-id Service
Clinical Data Sources: Discrete,
Text, ImagesIndexing Service Indexes
Honest Broker
Users
De-identified Data in DMZ
NLP Service
Federated Authentication
and Data Movement
Secure Data
Access Service
ETL Process
HL7 Service
Structured Data
Service
Custom Import Service
WSI De-id
Annotations
Cen
traliz
ed A
utho
rizat
ion
Ser
vice
Pub
lish/
Sub
scrib
e S
ervi
ces
Image Import Service
Penn
RPCI
Augusta
Pitt
Stony Brook
Jefferson
TCRN Network + TCIAIdentified Data Behind HIPAA Firewall
Text Annotation
Service
Feature Extraction Pipelines
QuI
PTH
RIV
E
Ann
otat
ion
Ser
vice
Dat
a S
ervi
ce
Sea
rch
Ser
vice
Pub
lic D
ata
TCIA
Figure 3. TCRN technical architecture enabling intra- and inter-communication of components using wrapping with microservices.
Table 2. Core Services in TIES Architecture
Service Tiers Description Services
Data Acquisition
Services
A set of services responsible for importing structured, WSI, and textual data; de-id service for managing linked identified/de-identified data; collection transfer for pushing HIPAA-compliant data to research data
● HL7 ● Structured Data ● Custom Import ● Image Import ● Indexing ● NLP ● Text Annotation ● De-Identification ● WSI De-identification
Data Retrieval Services
A set of services that enable data discovery, selection, and extraction of feature annotations
● Search ● WSI Annotation ● Data
Network Services
A set of services to provide federated identity management with single sign-on, authorization capabilities, manage publish/subscribe access and file transfer
● Centralized Authorization ● Publish/Subscribe