caBIG In Vivo Imaging Breakout caBIG Annual Meeting April 10, 2006

1
caBIG In Vivo Imaging Breakout caBIG Annual
Meeting April 10, 2006
2
caBIG Annual Meeting In Vivo Imaging
BreakoutIntroduction by Eliot Siegel, MDIn Vivo
Imaging Workspace Lead

• Formation of the caBIG Imaging workspace somewhat
  controversial last year
• Interesting mismatch between clinical use of
  imaging in the hospital and outpatient setting
  and the use of imaging in clinical trials, why is
  that the case?
• Difficulty getting images from various sites
  conducting clinical trials in comparison to text
  based and other data
• Lack of optimal quantitative tools to evaluate
  change in tumor over time on imaging studies
• Diagnostic imaging is and will continue to play
  an increasingly critical role as a biomarker for
  disease from both a clinical and research
  perspective

3
Introduction

• Unfortunately there are still major problems with
  images related to clinical trials
• Each clinical trial group has had to reinvent the
  wheel which is inefficient and very expensive
• ACRIN
• QARC
• BIRN
• others
• Current information systems such as PACS and the
  EMR are oriented toward clinical care rather than
  research

4
Introduction

• Imaging Informatics is an emerging discipline in
  medicine and will be increasingly important in
  the future of cancer care
• We believe that this field of informatics holds
  the key to making images accessible and also
  putting a greater degree of rigor on quantitative
  assessment of change over time which is critical
  fo the success of imaging as a biomarker
• Close working relationship between NCICB
  (NCI-Center for Bioinformatics) NCIA (National
  Cancer Imaging Archive) and Imaging Workspace

5
caBIG Annual Meeting In Vivo Imaging
BreakoutIntroduction by Eliot Siegel, MDIn Vivo
Imaging Workspace Lead

• Goals of the workspace
• Advance imaging informatics on multiple fronts
• Create, optimize, and validate software tools
• E.g. extensible imaging platform
• Model methods to extract meaning from in vivo
  imaging data and establish databases to test and
  validate these methods

6
In Vivo Imaging Workspace

• Structure
• Eliot Siegel Workspace Lead
• Four Special Interest Groups (SIGs)
• Software
• Standards and Interoperability
• Testbed
• Vocabularies Common Data Elements.
• Fifteen Funded Subject Matter Experts (SMEs)
• More than Seventy Volunteer Participants
  including participants from industry
• Participation of NCIA and NCI/CB
• Small Animal In Vivo Imaging

7
caBIG Annual Meeting In Vivo Imaging
BreakoutVocabularies and Common Data Elements
Special Interest GroupLead by Curt Langlotz,
MD, PhD and Daniel Rubin, MD

• Imaging VCDE SIG Goals
• Promote, support, develop, and evaluate
  standards-based vocabularies, ontologies, and
  CDEs for radiology and allied imaging fields.
• Participate in the design of the testbed and
  provide the vocabulary-related elements required
  by the testbed .
• Help develop the standards for creating, storing,
  and retrieving image metadata and image
  annotations.
• Harmonize VCDEs developed in the VCDE SIG with
  those being created by the VCDE workspace, and
  will develop VCDE-specific tools and resources
  that can be deployed on the grid to help realize
  the strategic vision of the caBIG effort.

8
Outline

• Background
• caBIG history
• Terminology for imaging
• ACRIN and imaging clinical trials
• Proposed imaging VCDE projects
• Structured image annotation and query software
• Terminology/CDE development for imaging
• Natural language processing

9
NCI Informatics Long Range Planning, circa 1999
The CII
10
Importance of Common Data Collection Methods,
circa 1999

• Serve as building blocks for the CII
• Allow pooling of data and comparison of results
  among clinical trials
• Facilitate enrollment of patients in clinical
  trials
• Avoid redundant data collection (capture-once,
  use-many times principle)
• Automate and expedite administration of clinical
  trials

11
Medical Vocabularies Completeness for Radiology
Langlotz Caldwell, J Digit Imaging 15(1S)201,
2002
12
What is RadLex?

• Sponsored by the Radiological Society of North
  America (RSNA)
• 26 participating organizations
• 9 committees
• 92 radiologist participants
• 5,308 anatomic concepts

10-30 percent of these concepts are not found in
SNOMED-CT
13
Lexicon Development Process
14
BOOP Search
mirc.rsna.org/radlex/service
15
American College of Radiology Imaging
Network(ACRIN)

• NCI-funded imaging clinical trial cooperative
  group
• Dozens of trials funded, including some very high
  profile trials (DMIST, NLST)
• Tens of thousands of subjects
• Case report forms containing thousands of
  potential CDEs

16
Proposed Imaging VCDE Projects

• Structured image annotation and query (IRW)
• Image meta-data standards
• Image annotation and structured data capture
• Image query by content from annotated image
  database
• Data collection methods for imaging clinical
  trials, harmonized to RadLex and caDSR/EVS
• ACRIN data collection elements
• DICOM elements
• The imaging playbook Cancer imaging devices,
  procedures and protocols
• Natural language processing (NLP)
• Evaluation of existing tools
• Adaptation or development of tools for radiology
  images

17
Image Annotation and Structured Data Capture
Capture data once, use it many times
18
Data Collection CDE Example

• Please describe the margins of the mass
• Smooth
• Lobulated
• Irregular
• Spiculated
• Obscured

19
Data Collection CDE Example

• Please describe the margins of the mass
• Smooth
• Lobulated
• Irregular
• Spiculated
• Obscured

20
Reusable Common Data Elements (CDEs) for Imaging

• Create caDSR-compatible CDEs from ACRIN data
  collection methods
• Identify CDEs specific to cancer imaging research
  needs
• Compliant with caDSR, harmonized with RadLex and
  EVS
• Associate atoms (terms) and molecules (CDEs)
• Move from lexicon (lists) toward ontology
  (knowledge)
• Coordinate with caBIG VCDE Workspace

21
The Playbook for Imaging in Cancer Research

• Vocabulary for imaging devices, procedures, and
  protocols
• (e.g., 7T 18-cm horizontal bore 4.7T 33-cm bore
  magnet operating at 200 MHz for 1-H imaging
  experiments)
• Common, vendor-independent language to describe
  experimental imaging instruments.
• (e.g., fast spin echo vs. turbo spin echo MRI
  sequence)

22
Natural Language Processing

• Unstructured information will always exist
• Narrative radiology report archives
• Peer-reviewed literature
• Focused extraction from radiology report
• Anatomy, findings (e.g, nodules and their
  descriptors), change over time
• Automatic population of reporting templates
• Inventory existing NLP tools
• Select or develop NLP tools to fulfill
  requirements

23
Vocabulary/CDE Strategy
Metadata storage formats
NLP
Metadata for Images
Image Annotation
Terminologies CDEs
Queries Analysis
Data Capture Formats Tools
Vocabularies CDEs
Data Re-Use Applications
24
Standards and Interoperability Special Interest
GroupLead by David Channin, MD and Paul Nagy, PhD

• Why Standards?
• Image Standards
• Workflow Standards
• Annotation Standards

The great thing about standards is that there
are so many to choose from. Dr. Andrew
Tanenbaum
25
Why Standards?

• Today, mountains of image data from clinical
  trials lies fallow.
• The appropriate use of standards can allow re use
  of the image data for other purposes than the one
  immediate trial.
• Thus enabling discovery in unanticipated ways.
• Computer Assisted Diagnosis
• Content Basis Image Retrieval

26
Image Standards

• Clinical Standard Medical Images come in
• DICOM (Digital Communications in Medicine)
• Loads of meta data
• Imaging Physics
• Frame of reference
• Patient/Study Information
• Naming inconsistent for re use
• Working with UPICT
• Uniform Protocols in Clinical Trials
• http//www.upict.org
• RSNA, FDA, NCI, AAPM, ..
• Mapping to VCDE (Vocabulary)

27
Workflow Standards

• How do we extract the image data from clinical
  environments?
• Not a great deal of technical onsite expertise
• Anonymization of PHI (Pseudonymization)
• Electronic submission to a repository
• How do we expose the data to researchers
• Query of meta data
• API autonomous access
• Goal is to allow interoperability at multiple
  layers in the technology stack of the Image
  Platform.

28
Workflow Comm/Query Standards
IHE Radiology

• LAN Based DICOM Q/R C-Store GPWL
• Internet based IHE RHIO Registry/Repository
  using EbXML/SOAP
• Internet - DICOM WG23 utilizing OGSA

29
Standards and Interoperability Special Interest
GroupLead by David Channin, MD and Paul Nagy, PhD

• Annotations and Image Markup In conjunction
  with Vocabulary

Courtesy Dr. David Clunie
30
Annotations and Markup

• Goal is to create a knowledge representation
  (OWL) for annotations in markup to enable
  semantic web applications.
• Provide practical presentation states in DICOM
  Structured Reports and XML RIDER.
• Create tools in the Imaging Platform to author
  this markup.

31
Imaging Software SIG Goals and Objectives

• The goal of the Software SIG is to create and
  adapt open source software tools to promote and
  enhance the use of imaging in cancer research.
  The SIG will focus on tools for image
  acquisition, management and analysis for use in
  clinical trials.
• specifically tools for enhancing lesion
  detection, characterization and change
  determination.
• The SIG will define requirements for these
  projects and write requirements specifications
  and/or white papers.
• The SIG will define use cases and test plans for
  each project and guide and track the development
  team that is tasked with implementation.
• The SIG will participate in the validation of
  software and/or algorithms resulting from each
  project using the IVI test bed

32
Viewing, Annotation and Analysis Software

• To facilitate the increased use of imaging based
  end points in clinical trials the SIG has
  identified the need for an easily extensible open
  source platform to support image analysis and
  visualization.
• To address this need a development program will
  been undertaken to create an eXtensible Imaging
  Platform (XIP)

33
eXtensible Imaging Platform

• The XIP is a
• Collection of software classes, algorithms and
  sample applications for building imaging
  applications valuable to research
• Method for rapidly prototyping "medical imaging
  workstation" applications from a re-usable,
  extensible set of modular elements
• Researchers will be able to rapidly develop and
  evaluate new approaches to medical imaging
  problems, and use them in a translational
  research setting.
• Grid technology in general, and caGrid in
  particular, makes it possible to let users to
  choose between grid components and locally
  available components.
• Analytic services (CAD algorithms, algorithms for
  quantifying changes in consecutive imaging
  studies, algorithms associated with a 3-D
  visualization pipeline etc).
• Data sources might or might not be DICOM based.
• Both data and algorithms can be physically
  distributed.

34
Current Status

• Imaging Software SIG has developed a requirements
  specification for the XIP
• An RFP has been drafted
• The SIG is working to define appropriate
  milestones and demonstration projects

35
Change Detection Analysis

• The In Vivo Imaging Workspace is assessing
  current status of change detection analysis
  technology for cancer imaging
• Detecting and quantifying change in lesions over
  time represents a critical unmet need in the
  Cancer Research Community

36
Baseline
Follow-up
37
Baseline
Follow-up
38
Change Analysis and Validation

• Working on definition of SIGs role in larger NCI
  activity. It has been suggested by NCI that a
  major contribution would be the development of
  basic change analysis algorithms, and evaluation
  methods.
• Algorithms for binary outcome determination and
  for change quantification
• Databases with known truth for validation studies
• Databases containing multiple segmentation
  results on the same images using different
  approaches
• Utilize the plug-in application interface for
  the XIP to provide a sand box in which
  algorithms may be implemented and evaluated

39
Testbed Special Interest GroupLead by Joel
Saltz, MD, PhD and Stephan Erberich PhD

• SIG Goals
• Design and implement core middleware compliant
  with caGrid, DICOM and IHE
• Addresses the need for high performance data
  transport on the grid, and dynamic algorithm
  deployment to reduce the need to data movement.
• Develop software development environments to help
  developers use middleware to develop applications
• Work with cooperative groups to leverage testbed
  capabilities in support of translational research
• Responsibility for coordination of GridCad
  application

40
Testbed Special Interest GroupLead by Joel
Saltz, MD, PhD and Stephan Erberich PhD

• Testbed Consists of vivo imaging caGrid
  standards, reference middleware stack
  implementation supporting grid based
  applications. The testbed is designed to support
  each individual application as well as to
  demonstrate interoperability between
  applications
• Testbed Projects Middleware, Coordination of
  Application Projects, Cooperative group outreach

41
Middleware Testbed for multi-center clinical
trialsCooperative Cancer Group use case
scenario

• ACQUISITION
• Image acquisition and handling at trial site
  (Image transfer techniques, HIPAA, firewalls,
  MIRC)
• Quality assurance
• REVIEW AND ANALYSIS
• Image Warehousing, access control, and central
  review
• Access to correlative and image meta data via
  caGrid
• Annotation and Markup in caGrid
• Quantitative analytic tools
• DISCOVERY
• Integrated caGrid supported discovery of image,
  molecular, pathology information

42
Testbed development focus application gridCAD
A Novel Use of Grid Computing to Support Human
Markup and Execution of Multiple CAD Systems

• Tony Pan, Joel Saltz, Tahsin Kurc, Stephen
  Langella,
• Shannon Hastings, Scott Oster, Ashish Sharma,
  Metin Gurcan
• Department of Biomedical Informatics
• The Ohio State University Medical Center,
  Columbus OH
• Eliot Siegel, Khan M. Siddiqui
• University of Maryland School of Medicine,
  Baltimore, MD

43
Benefits

• Facilitate research and clinical decision support
  with large number of subjects and multiple CAD
  algorithms.
• Parameter studies, clinical and preclinical
  trials, etc
• Provide a client to support remote human markup
  of nodules
• Enable better algorithm development and
  validation through the use of many distributed,
  shared image datasets
• Support remote algorithm execution reduce data
  transfer and avoid the need to transmit PHI
• Reduce overall processing time and algorithm
  development cycle through remote compute resource
  recruitment and CAD compute farms
• Scalable and open source caGrid 1.0

44
gridCAD Architecture
Expose algorithms, human markup and image data
as caGrid Services
45
Future Direction

• Location independence
• Move algorithms to data
• Move data to algorithms
• Move both data and algorithms to compute servers
• Currently supported ongoing collaborations to
  deploy these capabilities
• Security and Privacy
• Encryption and Just-In-Time anonymization for the
  image data services
• Scaling and Deployment
• High performance image transfer mechanisms
• Greater number and variety of CAD vendors
• Additional application areas, including CAD for
  other diseases and in vitro image
  analysis

46
COG/NANT Cooperative Group Application

• COG Phase-I Consortium (23 medical center) and
  NANT (14 medical centers) are now actively
  engaged in the caBIG testbed.
• Grid based analysis of perfusion imaging studies
  DCE-MRI analysis deployed as an analytic service
• Grid based evaluation of joint prognostic value
  of perfusion studies, pathology, molecular
  clinical data

47
In Vivo Imaging Workspace involvement in RSNA
2006Eliot Siegel, MD,
6 Workstations SIG 1 Testbed Architecture
3 Workstations SIG 3 Vocabulary
2 Workstations caBIG demo 1 spy, Rembrandt
2 Workstations NCI CIP demo Projects IDRi
Theme Park Directory Purpose
6 Workstations SIG 2 Software
3 Workstations SIG 4 Standards
2 Workstations NCIA RIDER
4 Workstations Allies Pharma Device ?CRO
48
QA and Wrap UpLead by Eliot Siegel, MD, In Vivo
Imaging Workspace Lead

• Engage in projects that further the strategic
  goals of the Imaging Workspace and caBIG
  program.
• Identify synergies with the other caBIG
  workspaces.
• Partner with external organizations within the
  caBIG community, (ex. ACRIN, NCIA), to further
  Imaging Workspace and caBIG program goals