Frank A' Voelker, DVM, DACVP

1
Practical Image Analysis from a Pathologists
Perspective

• Frank A. Voelker, DVM, DACVP
• Pathology Experts LLC

2
Topics.

• Introduction
• General Concepts and Approaches
• Guidelines and Pitfalls
• Analytical Strategies
• Applications and using Genie
• Summary

3
General Analytical Approaches.
Pixel Count
IHC Deconvolution
Co-localization
IHC Nuclear
Rare Event
Membrane
Angiogenesis
4
Two Different Approaches for Analysis
Quantify Histomorphologic Change

• Cellular Hypertrophy/Atrophy
• Cell Numbers
• Tissue Infiltrates (eg. Fibrosis)
• Other Structural Alterations

Usually measuring area or number
Quantify Substances using Special Stains
Usually measuring area and/or intensity

• Histochemistry
• IHC
• ISH

5
pS6 Ser235 Immunostain of Breast Carcinoma
Introducing the Concept of Targeted Cell
Analysis
Analysis of average cytoplasmic stain intensity
using the pixel count tool may be useful in
evaluating a neoplasm if there is little
background or nonspecific staining.
6
Fibrosis in Livers of Zucker Rats
Use of the Positive Pixel Count Tool enables
visually apparent analysis of a change
T
T
Control Rat No. 12
Fenofibrate Rat No. 5
T
T
C
Pioglitazone Rat No. 3
Compound X Rat No 2
C
F
P
X
Variations in fibrosis (blue) about small portal
triad veins (T) as depicted using Massons
Trichrome stain
7
Quantitation of PAS Stain for Glycogen in Livers
of DIO Mice Administered XXX Using the Aperio
Color Deconvolution Tool
Using the Color Deconvolution Tool enables
quantitation of things visually obscured by
counterstaining

PAS-stained Section
Aperio Markup Image
8
Cyclin D1 Immunostain of Human Breast Carcinoma
Use of the IHC Nuclear Analysis Tool to Determine
Percent and Degree of Positivity of Neoplastic
Cell Nuclei. Stromal Nuclei are Excluded from
Evaluation.
9
Quantifying Inflammation in Tissue using the
Nuclear Analysis Tool
Different cell types often can be distinguished
from each other in the same tissue based on
nuclear diameter. Here lymphocyte nuclei are
smaller than mammary carcinoma nuclei.
This makes it possible to count lymphocyte
numbers per unit area of tissue cross section to
determine degree of infiltration.
Algorithm IHC Nuclear (cell-based)
10
Mouse Liver - Hepatocellular Hypertrophy
Drug-related enzyme induction leading to
increases in cytoplasmic endoplasmic reticulum
with resultant hepatocyte size increase.
Total Hepatocyte Nuclei 199 Average Nuclear
Size 140 µm² 706 nuclei/mm²
Total Hepatocyte Nuclei 167 Average Nuclear
Size 160 µm² 508 nuclei/mm²
Algorithm IHC Nuclear (cell-based)
11
Some Guidelines for Analysis of Slides from
Experimental Studies

• Take care to assure immediate optimal fixation
  for all tissue samples. Uniformity of handling
  as well as fixation time is important.
• Staining procedures for all slides in a study
  need to be performed simultaneously in a single
  batch to assure uniformity of stain.
• Sampling must be strictly representational as
  well as consistent. Care must be taken to assure
  exact uniformity of analysis with respect to
  anatomical location (eg. Tissue trimming,
  sectioning)
• A preliminary evaluation of image analysis tools
  between some slides of varying stain intensities
  will help assure that analysis values are
  established optimally for all slides in the
  study.

12
Anatomic Consistency in Sampling..
13
Sirius Red Stain Depicting Myocardial Fibrosis in
a Mouse
Analysis Tool Color Deconvolution (area-based)
Precision in level of section is required for
accurately comparing amounts of fibrosis between
treatment groups
14
Consistency of Sample Area Selection for
Morphometric Analysis within the Median Lobe of
the Mouse Liver
1
2
3
Select samples within approximately the same
region of the same lobe of the liver for
consistency of analysis. As an assurance of
sampling homogeneity, areas should have roughly
similar pixel count values.
15
Consistency of Study Conditions can Affect
Morphometric Analysis
Variations in duration of fasting prior to
necropsy can result in large differences in
hepatocellular glycogen thus leading to
inaccurate analysis
263 nuclei/mm²
212 nuclei/mm²
Mouse Livers
16
Three Possible Strategies for Measuring Brown
Stains using the Positive Pixel Count Analysis
Tool

• Quantitate the percentage area of all brown
  pixels in the section or in selected areas of the
  section.
• If the chromagen staining is very extensive in
  the target cell population, measure only the
  brownest (darker) pixels in selected areas of the
  section.
• If the chromagen staining is uniform in character
  and very extensive in the target cell population,
  measure stain intensity as an index of
  concentration.

17
Percent of Liver Tissue Staining for Transferrin
Receptor(CD71) in Female Mice by
Immunohistochemistry
Measuring all of the brown pixels in the sample
area



Control
100 mg/kg
250 mg/kg
1000 mg/kg
p ? .01 p ? .001
18
Quantitation of Cytochrome p450 Reductase in
Centrilobular Hepatocytes Despite Widespread
Immunostaining
Original Image
Markup Image
Measuring only the area of more intense stain
Color deconvolution (area-based)
19
Quantitation of VEGF Immunostaining in Livers of
Mice administered XXX for 52 Weeks
Control Females
Control Males
1000 mg/kg Males
1000 mg/kg Females
Comparing stain intensity
20

The Challenge of Analyzing only the Target
Tissue.
PTEN Immunostain of Squamous Cell Carcinoma in
Human Lung
Variable staining of neoplasm and staining of
surrounding stroma make morphometric analysis
difficult.
21
Automated Recognition of Neoplastic Components in
a Human Bronchoalveolar Carcinoma using Genie
Recognition of neoplastic tissue components
within a neoplasm is an important first step in
quantifying amounts and intensities of specific
biomarkers using IHC. This is needed for
accurate clinical trial assessment of
antineoplastic agents
The next step would be to link neoplastic tissue
recognition using Genie with a color
deconvolution tool for measurement of chromagen
in an IHC stain.
22
Genie..
Introducing the concept of using histology
pattern recognition software as a preprocessing
machine to segregate target from nontarget tissue
during analysis
Strategies
23
Steps in Chromagen Analysis of a Neoplasm
(Excluding the Stroma)
Primary IHC image
Geniemarkup with selection of neoplasm
1
2
Eliminate stroma
Final Aperio ImageScope deconvolution markup
4
3
24
Quantitation of Splenic Extramedullary
Hematopoiesis in a Mouse using Genie and the
Aperio Positive Pixel Count Tool
Genie Markup Image
HE Stain
Positive Pixel Markup Image
Results EMH comprises 50.2 positive pixels in
evaluation area
25
Quantitation of Periarteriolar Lymphoid Tissue in
a Mouse Spleen using Genie and the Aperio
Positive Pixel Count Tool
HE Stain
Genie Markup Image
Aperio Positive Pixel Markup Image
Result Lymphoid tissue comprises 30.1 of
positive pixels in splenic cross-sectional area
Extrapolating to an entire tissue section demands
more robust training than for a simple image.
26
Analysis of Study Sample Groups by Genie
Morphologically Variable Samples Trained
Individually for Genie Target Tissue Selection
Targeted Tissue Selection and Isolation by Genie
Subsequent Uniform Analysis of Isolated Target
Tissue for area/intensity
Separate target tissue training of each sample
does not adversely affect final analysis.
27
Bile Duct Hyperplasia in Rat Liver
First pass Genie histology pattern identification
with minimal training. Genie can simultaneously
analyze three or more tissue areas
Hyperplastic Bile Ducts Green Hepatic
Parenchyma Red Periportal Inflammatory Cells
Blue Periductal Collagen Brown Bile Duct Lumena
Sinusoids Yellow
Then analyze up to three tissue areas using
colocalization tool
28
Quantitation of Hepatocellular Necrosis
Use of Genie as a preprocessing utility to
identify regions of hepatic necrosis (red) and
areas of normal liver (green)
Subsequent quantitation of necrotic areas using
a pixel count tool to allow precise grading
29
Using Genie to Discriminate Between Nuclear and
Cytoplasmic Markers
Human Breast Carcinoma Stained for Estrogen
Receptor
The ability of Geni to discriminate between
nuclear and cytoplasmic regions of a neoplasm
allows separate biomarker intensity measurement
for both nuclear and cytoplasmic markers.
30
Monkey Lung
Use of Genie as a preprocessing utility to
identify regions of smooth muscle (green)
Subsequent quantitation of pulmonary smooth
muscle using a pixel count tool
31
Cynomolgus Monkey Lung
Use of Genie as a preprocessing utility to
identify regions of bronchiolar epithelium
(green)
Subsequent isolation and analysis of only
bronchiolar epithelium using the positive pixel
count or other analysis tool
32
Islet Cell Mass of Mouse Pancreas
Measurement of Pancreatic Islet Cell Mass using
Genie Followed by the Colocalization Algorithm
(A/B)CIslet Cell Mass
ATotal Islet Area in Section
BTotal Pancreas Area in Section
CPancreatic Weight
33
Estimating Islet Cell Hypertrophy in the Mouse
Pancreas
Calculating Mean Islet Cell Area using Genie
followed by the IHC Nuclear Algorithm
Total islet area 103014 µm² Total number
islet nuclei 575 103014 µm²/575 179 µm²/islet
cell
34
Quantitating Dog Thyroid Gland Tissue Components
Use of Genie as a preprocessing utility to
identify thyroid gland follicular epithelium
(green), colloid (red) and C-cells (blue)
Then quantitate each separate tissue component
area using the colocalization tool.
35
Measuring Cellular Hypertrophy of two cell types
in a Dog Thyroid Gland
Set Genie masks for brown follicular epithelium
and blue c-cells. Then apply colocalization tool
to calculate respective areas of each.
Then apply IHC nuclear tool on same image to get
numbers of artificially colored brown and blue
nuclei.

Total Brown Area/Total Brown Nuclei Mean
Follicular Cell Area. Do same calculation for
blue nuclei.
36
Summary

• The ability to digitize entire slides and perform
  morphometric analysis on images has been
  valuable in allowing the rapid and practical
  measurement of tissue biomarkers for
  pharmaceutical research and development.
• A number of strategies and examples have been
  presented for using various image analysis
  algorithms in the measurement of tissue changes
  and tissue biomarkers. Image analysis of
  specific target tissues can be particularly
  challenging in cases with large and
  morphologically intricate areas of tissue, or
  when tissue staining is nonspecific.
• Genie, a histology pattern recognition tool,
  has been introduced as a preprocessing utility
  capable of identifying and categorizing specific
  histologic tissue types, thus allowing subsequent
  analysis of target regions by standard image
  analysis tools.
• Significant challenges remain in developing
  practical procedures and methods appropriate for
  the analysis of oncology and toxicology
  specimens. Recent object recognition
  advancements may assist in this effort.

37
Acknowledgements

• Ms. Kimberly Merriam, TBG, BMD Novartis
• Ms. Jeanette Rheinhardt, TBG, BMD Novartis
• Dr. Allen Olson, Aperio
• Dr. Kate Lillard-Wetherell
• Mr. James Deeds, Oncology Research Novartis
• Dr. Rudi Bao, Oncology Research Novartis
• Dr. Humphrey Gardner, TBG, BMD Novartis
• Dr. Alokesh Duttaroy, DMDA Novartis
• Dr. Steve Potts, Aperio
• Dr. Reginald Valdez, Novartis
• Dr Oliver Turner, Novartis
• Many Others

38
Frank VoelkerDVM  MS Diplomate ACVPKey bio
points / specialties
Pathology Experts LLC provides its sponsors with
the highest level of expertise in toxicologic
pathology with focus on adding value to the
preclinical phase of drug and device development.
We provide a wide range of consulting services
across all major therapeutic areas and organ
systems. Contact us _at_ www.pathexperts.com Basel
Switzerland Rye, New York USA