Digital Pathology Image Analysis in Pharmaceutical Discovery and Development different uses, differe

1
Digital Pathology / Image Analysis in
Pharmaceutical Discovery and Development-
different uses, different concerns

• Daniel Weinstock DVM PhD DACVP
• sanofi aventis U.S., Inc.
• Bridgewater, N.J. USA

2
The Digital Image Revolution

• Histopathologic assessment (the traditional
  method)
• - glass slides and an optical
  microscope
• - subjective semi-quantitative
  assessment by a pathologist with
  peer review of results
• New approach
• - digital image acquisition with
  computer based image
  handling and viewing
• - pathologist driven analysis with
  generation of objective
  quantitative, data
• (Why) is this such a good thing?

3
Pathology Applications in a Pharmaceutical Company

• Discovery (Research)
• Target Validation
• High Content Screening (HCS)
• Animal models
• Proof of concept / proof of mechanism studies
• Development
• GLP toxicologic pathology
• Biomarker development / validation
• Investigational toxicologic pathology

4
Image AnalysisWhat kinds of questions?

• Characterization of changes in cells / tissues
• what kinds of changes
• severity and distribution
• Frequency and distribution of a microscopic
  feature
• normal versus diseased
• treated versus untreated
• Challenges
• non-uniformity of samples
• Variations in sample source, handling and
  staining
• small sample numbers
• large sample numbers
• subtlety of change
• spectrum of change

5
Digital Imaging and Image AnalysisApplications,
Concerns and Reasons for Use

• Repeated measures
• uniform analysis (application of algorithm)
• Quantitative analysis
• hard numbers for diverse scientists
  (committee decisions)
• Large sample numbers
• prevents drift
• Distance pathology / telepathology
• remote image sharing
• collaboration / consultation
• GLP principles image handling, storage and
  archive

6
Digital ImagesAcceptance by Pathologists
Ultimate goal replace glass slide evaluation
via microscope with digital image evaluation on
computer screens
These issues must be addressed to the
satisfaction of the primary users of the
technology. Very good progress to date, but
improvement possible.

• Quality
• images
• data
• Speed
• slide scanning
• image access, handling
• field of view, magnification change, etc
• Cost and benefit
• Integration
• Ease of use

7
Image Analysis Practical Aspects

• Team approach needed
• fusion of engineering and biological skill sets
• statisticians needed for complex analytical
  techniques
• Criteria for evaluation
• modifiable algorithm until final parameters
  established
• Reiterative evaluation and modification of
  algorithm required
• Should be able to review results of each
  modification
• Repeated modification should yield incremental
  improvements in discrimination
• final application of unchangeable algorithm to
  total image set
• End point believable, repeatable, biologically
  relevant results
• e.g. recognition of a nucleus many
  ways to do it

8
Image Analysis How To

• Digital image files acquired and stored
• working algorithm applied
• 1st round results generated
• can apply to smaller representative image set
• evaluation of results and assessment of
  discrimination
• algorithm modification, data set exclusion
• Application of 2nd, 3rd, etc modified
  algorithms
• reiterative cycle of modification and data
  assessment
• Final data generation and analysis
• applied to total set of images
• final review of analyzed images for QA is
  desirable

9
Discovery versus Development

• Types of questions
• therapeutic effects (discovery) versus
  toxicologic effects (development)
• disease status, model and assay development
  (discovery)
• Types of tissues / experiments
• species differences
• Standard toxicology species (rat, dog, etc.)
  versus mice (genetically modified, knock-outs,
  knock-downs, etc.) and other species
• group size constraints
• reagent concerns
• Clients (end user)
• regulatory oriented (development) versus diverse
  scientific community (discovery)
• GLP compliance
• essential in Development, not relevant in
  Discovery
• Investigational Toxicologic Pathology hybrid
  between the two

10
(No Transcript)
11
Image Analysis Concerns Tissues

• Liver (example tissue)
• Multiple types of changes possible
• Variable combinations of changes separate,
  intermixed, etc.
• Range of severity of each type of change
• Necrosis
• Fibrosis
• Inflammation
• Bile duct proliferation
• others
• Normal features difficult to differentiate
• Red blood cells
• Sinusoids amount of space affected by degree of
  exsanguination
• Kupffer cell nuclei difficult to discern from
  inflammatory cells

12
Range of Changes in a Lesion

• Liver
• - necrosis
• Issues
• Red cells within area of necrosis
• Clear spaces within necrosis vs. sinusoids
• Pyknotic nuclei vs. Kupffer cell nuclei

13
Range of Changes in a Lesion

• Liver
• - bile duct proliferation
• Issues
• Edge effect.
• Differentiation between bile ducts and
  arterioles.
• Relatively uncomplicated change in this field.

14
Range of Changes in a Lesion

• Liver
• - bile duct proliferation
• - fibrosis
• - inflammation
• Issues
• Differentiation between bile ducts and
  arterioles.
• Complicated by fibrosis and inflammation.
• Discrimination between nuclei of inflammatory
  cells and Kupffer cells

15
Range of Changes in a Lesion

• Liver
• - bile duct proliferation
• - fibrosis
• - inflammation
• Issues
• Complex morphology of multiple changes in one
  focus of interest.
• Severity change varies by focus.

16
Range of Changes in a Lesion

• Liver
• - necrosis
• - fibrosis
• - bile duct proliferation
• - inflammation
• Issues
• Similar issues as previous images, but now
  complicated by multiple contiguous types of
  changes per field.

17
Range of Changes in a Lesion

• Liver
• - necrosis
• - bile duct proliferation
• Issues
• Multiple non contiguous changes.
• Bile duct proliferation uncomplicated.
• Necrosis complex morphology in area of change.

18
Image Analysis How to and Multiple
interactions

• Whats needed?
• turn key library with many validated algorithms
• can be located distant or local
• useful as starting point for further
  modification
• tool box for modification
• should be local (desktop)
• should be user (pathologist / scientist)
  friendly
• easily modified with rapid, repeated application
  to a test data set
• format for easy review of results and assessment
  of discriminations being made
• data should be accessible for statistical
  analysis
• final results should be biologically relevant

19
FAQs common concerns

• What must be done to validate an image analysis
  algorithm?
• What justifies the time and effort investment to
  develop an image analysis algorithm?
• How predictive is a 2 dimensional slice of a
  tissue (histologic section) for quantification of
  an effect on a organ? How much sampling is
  required? What kind of sampling is required?
  Are we making appropriate comparisons?
• What is necessary to power the experiment
  appropriately?

20
Integration of Images and Data

• GLP or non-GLP
• Necessary to be able to associate images with
  blocks, tissues, animal identification,
  treatments, experiments, etc
• source information, interface with LIMS
  (Laboratory Information Management System)
• cross reference to lab books
• Necessary to be able to associate images with
  multiple analyses and results
• cross reference in reports
• interface with document generation programs
• Storage and retrieval of images and data
  IS/IT participation essential
• searchable (on how many and what criteria?)
• image quality / integrity
• Compression, storage space and location
• storage of primary image, annotated images,
  etc.
• Trade off amount of annotation vs. ease of use
  (data entry time)
• potential for retrospective analysis

21
Technical Needs

• Rapid, automated slide scanning
• Multiple formats
• brightfield
• fluorescence
• Rapid, seamless change between magnifications
• Depth perception, polarization?
• Volumetric determinations?
• Pathologist / scientist supervised computer self
  learning for image analysis

22
Other applications

• Digital Imaging
• - Telepathology - sharing of digital images
• Image Analysis - cellular to whole animal
• - HCS (High Content Screen)
• - Transgenic mice with in vivo light emission
  (e.g. luciferase)

23
Large Scale High Content Screeninge.g.
Anti-mitotics
What is the relevant measurement? Discrimination
parameters based on experimental observations
(data) with appropriate controls is essential.
- Parameters are often not intuitive. -
Results must be biologically relevant to
mechanism of action. Morphology varies with time,
dose, staining and mechanism of
action. Sophisticated approach with complex
analysis (re-analysis) is needed.
24
Image Analysis HCS special issues

• Large experiments
• up to 384 well plates
• very large screens, very large data sets
• Feature extractions what, how, etc
• Image compression current use and archive
• resources for data storage become important with
  time
• loss of image integrity with compression may be
  an issue especially for retrospective
  analysis
• Data normalization
• inherent variations within an experiment
• Data mining multivariate analysis
• need for sophisticated statistical analysis
  multiple possible methods
• team approach essential
• final biological relevance is essential

25
Whole animal in vivo Bioimaging

• Transgenic animal with luciferase reporter
• luciferase (enzyme) is produced in response to
  specified gene expression
• enzyme substrate given intravenously
• whole mouse is imaged for in vivo light emission
• tissue imaged ex vivo
• image analysis used to quantify gene expression
  based on light emission

Journal of Molecular Endocrinology (2005) 35,
293-304
26
Evolution of the Process

• Technology and applications are in infancy
• New, easier, less expensive technology required
  for widespread acceptance and use
• Current investigators will validate the
  technology for traditional applications
• Future investigators who evolve with the
  technology will likely be ones to define new,
  unorthodox, innovative applications

27
Summarywhat a pathologist wants / needs

• Digital Images
• Quality images
• Rapid manipulations
• Integrated systems
• Easy to use
• Image analysis
• Quality data
• Pathologist / scientist driven
• Reiterative process for refinement of criteria
• Easy to use
• You cant always get what you want
• - Rolling Stones, Hot Rocks,
  1964-1971

Consider the constraints of the individual
workplace.