Whole Slide Image Based Interpretation of Immunohistochemistry Stains in Challenging Prostate Needle Biopsies

1
Whole Slide Image Based Interpretation of
Immunohistochemistry Stains in Challenging
Prostate Needle Biopsies

• Jeffrey L Fine MD, Jonhan Ho MD, Yukako Yagi,
  Drazen Jukic MD PhD, John R Gilbertson MD,
  Sheldon I Bastacky MD, Dana M Grzybicki MD PhD,
  Leslie Anthony, Robb Wilson, and Anil V Parwani
  MD PhD

2
Objectives

• Review whole slide image landscape
• Present research project
• Discuss implications arising from the study

3
Whole Slide Image (WSI)

• Digital facsimile of an entire glass microscope
  slide that is viewed by virtual microscopy (VM)
  software
• WSI are also known as Virtual Slides or
  Digital Slides

4
(No Transcript)
5
The Landscape WSI Systems

• Systems are currently self contained
• Image acquisition, management, storage, and
  utilization (viewing and image analysis)
• Bar code capabilities limited to reading

6
Obvious Current Trends

• Decreasing cost for robots and storage
• Increasing speed for robots
• Raw capture speed
• Better shortcuts
• Involvement of traditional microscopy players
• Olympus, Zeiss, Nikon

7
Nascent Trends

• Vendor concern about workflow and integration
• How to slip a robot into an existing APLIS and
  histology workflow
• Digital pathology workstations
• Monitors (how many and how large)
• Display calibration

8
Clinically-Oriented ResearchWSI Clinical
Evaluation Group

• Core affiliated group
• 4 pathologists 1 fellow study coordinator data
  coordinators imaging technicians LIS personnel
• Additional pathologists, depending upon study
• Prior studies
• Quality Assurance
• Primary Diagnosis

9
Current Project

• Goal Validation of WSI technology for
  interpretation of immunohistochemistry (IHC)
  stains
• Why?
• UPMC has a centralized IHC laboratory that
  supports two academic hospitals
• Electronic distribution (via WSI) could decrease
  turn-around time for IHC stains
• Better patient care better service to clinicians
• Decreased healthcare cost (shorter length of
  stay?)
• WSI could permit automated image analysis of IHC

10
Traditional workflow

• Slides stained
• Slides sorted and gathered
• When a group of stains is complete they can be
  shipped to pathologist
• Slides packed and shipped (courier)
• Received slides are sorted (again) and
  distributed to pathologists

11
WSI workflow

1. Slides are stained
2. Stained slides are placed into a slide scanning
   robot which reads their bar codes and does the
   heavy lifting (naming of file copying of file to
   server etc.)
3. Pathologist views the slides directly over the
   internet
4. Glass slides catch up later (optional?)

12
Prostate Needle Biopsies

• Availability at UPMC Shadyside
• Small set of usual IHC stains
• p63 cytokeratin 903 racemase
• Typically signed out in an itemized fashion
• Detailed information about each part or block
• Very challenging IHC interpretation

13
Cytokeratin 903 immuno stain stains cytoplasm
of basal cells
14
p63 immuno stain stains nuclei of basal
cells (positive noninvasive)
15
racemase (aka AMACR) immuno stain stains
cytoplasm of glandular cells in prostate
16
Retrospective StudyPossible UPMC Environment

• Stage I
• Pathologist has glass HE which requires IHC
  staining for definitive diagnosis
• Stage II
• Pathologist receives WSI of IHC stains and
  interprets them
• Stage III
• Glass IHC stains are eventually received and are
  checked by the pathologist
• Consensus conferences

17
Study Design

• 100 cases screened
• 30 difficult foci found
• Each study case represents one focus

18
Technology

• High throughput WSI system
• T2 (Aperio Technologies, Vista, CA, USA)
• Viewing
• Either WWW-based viewer or standalone viewer
  (both supplied by WSI vendor)
• Standard desktop PCs and microscopes
• Server
• Nothing special (5 users and 17 20 GB)

19
Data Collection

• Stain by stain interpretation (stages 2 3)
• Overall Diagnosis
• Confidence in diagnosis
• Time required to make diagnosis (roughly)
• Complexity of case
• Quality of each slide or image
• Explanations for any defects or shortcomings,
  including network speed

20
Stain Interpretations

• Positive
• Negative
• Cant Tell (?)
• Subcategories to help determine why the
  pathologist couldnt intrepret the stain

21
Additional Data Collection

• Consensus Diagnosis
• Is mild disagreement OK (atypical vs. cancer)
• How did this compare with original diagnosis
• Any relevant features or notes about case
• Image defects (de-focused areas color
  reproduction etc.)
• Poor stain quality (not the images fault)

22
Results
23
Intra-observer Agreement(Stain Interpretation
WSI vs Glass)

• Five Pathologists
• Average Intra-observer agreement
• 80.6 (standard deviation 4.5)
• Range (75.7 - 86.0)

24
Additional ResultsImage Defects

• Pre-existing QC procedure did not detect several
  defective images
• Edge Defects
• Rotation Defects

25
Edge Defect
26
Rescanned
27
Rotation
HE
Immuno
28
(No Transcript)
29
(No Transcript)
30
(No Transcript)
31
Discussion
32
Validation

• Does this study validate WSI for interpretation
  of IHC stains?
• Pathologists agreed with themselves about 80 of
  the time
• Need to find most common sources of disagreement
  and see if they can be addressed
• It does highlight several points that need to be
  addressed prior to using WSI technology for real
  clinical applications

33
WSI Quality Control

• Each WSI must be checked for common defects
• This has to be automated eventually
• All slides are not equal
• IHC stains are susceptible to edge defects
• Frozen section slides are hard to get focused
• Image quality standards do not exist yet for WSI

34
Modification to WSI Process

• Created a QC procedure (manual)
• Includes solutions/fixes
• Performed by technical support staff
• Documentation of QC activities (aka QA)
• Log files
• Monitor image quality
• Minimize sub-optimal or defective WSI that are
  released to pathologists

35
Workflow

• Glass was felt to be faster
• Current pathology systems do not accommodate WSI
• Look up case in pathology system and click on
  available slides

36
(No Transcript)
37
(No Transcript)
38
Viewer Limitations(Most Systems)

• Image navigation
• (slow click and drag)
• cannot rotate image easily (GI skin IHC stains)
• Presentation speed is slow
• (pixels are visible until image can load
  completely)
• Lack of clinical data integration
• (whos slide is this?)

39
Study Flaws

• Pathologist subjects
• Informatics fellows non-GU pathologist
  GU-trained sub-specialists
• Almost all pathologists were informatics
  pathologists
• No standard display or VM software
• 2 options for VM software
• No gold standard for monitor/PC
• Loose track of time

40
Future Work

• Address flaws
• Pathologist selection
• Attention to software and computer used to
  participate in study
• Other applications
• Frozen Sections

41
Conclusions

• This study provided experience in the attempted
  production of clinical grade pathology images
• Experience has altered our QC procedures
• Further tools are needed (automation,
  integration, etc.)

42
Conclusions

• If validated (not yet), WSI technology could
  permit electronic distribution of IHC stains
• Reduced turn around time could improve service
  and reduce healthcare cost
• Centralized laboratories could support multiple
  hospitals or pathologist groups
• Automated image analysis could be a future source
  of added value

43
Conclusions

• WSI technology is entering a new phase
• Machines/systems are adequate for small scale
  educational and research use
• WSI systems are not yet capable of integration
  with existing pathology systems
• This study (when published) can stimulate vendors
  and mainstream pathologists effectively
  transition to the next level

44
Acknowledgements

• Rebecca Crowley MD
• Michael Becich MD PhD
• Russ Silowash
• Jon Duboy