what difference does a difference make?
DESCRIPTION
What difference does a difference make?. Elizabeth Little, Ph.D. 26-Oct- 2010. Talk overview. Introduction Tissue thickness variation Using best histological practices Stain intensity variation due to tissue thickness The difference matters Could impact algorithm functionality. - PowerPoint PPT PresentationTRANSCRIPT
What difference does a difference make?Elizabeth Little, Ph.D. 26-Oct- 2010
Talk overview
• Introduction
• Tissue thickness variation– Using best histological practices
• Stain intensity variation due to tissue thickness
• The difference matters– Could impact algorithm functionality
Systems integration
source: www.vagabondish.com
The Hematoxylin & Eosin (H&E) slide
• Numbers– In 2009, 330 million histology slides were produced in the United
States– 83% (274 million) were stained with H&E
• Pathologist– Potential first look at the disease state
• Cost– Dollars vs. thousands of dollars for more advanced testing
Impacts of H&E stain variability
• Pathologist workflow is impacted by staining variability– Repeat slides
• Imaging workflow is also impacted by staining variability
– Algorithms can by impacted by stain variability
Antecedents that are helpful for H&E slide image analysis
• Control of the stain variation– Under best practices we can control stain variability to a certain
degree
• Algorithms that are robust against stain variation
Staining variables we cannot control- tissue type affects stain intensity
0501001502002500
2000
4000
6000
8000
10000
Grey scale intensity differences - skin vs. kidney
Kidney
Skin
Intensity Level
Pixelcount(N)
Staining variables that we have some control over - tissue thickness impacts stain intensity
2 micron 4 micron
255 204 153 102 51 00
1000
2000
3000
4000
5000
Intensity level
Grey scale intensity difference due to tissue thickness
2 micron slice
4 micron slice
Pixelcount(N)
Talk overview
• Introduction
• Tissue thickness variation– Using best histological practices
• Stain intensity variation due to tissue thickness
• The difference matters– Could impact algorithm functionality
Possible sources of variations in section thickness in the histology laboratory
• Fixative
• Duration of fixation
• Tissue processing
• Paraffin
• Tissue block
• Microtome
• Histologist
Objective – measure the sectioning process impact on tissue thickness
• 1 tissue block used
• 1 microtome
• 2 settings– Automated (32 slides per histologist)– Manual (32 slides per histologist)
• 2 histologists– 22 years of experience vs. 4 years of experience
Tissue thickness variability testing outline
• Section – Tissue was sectioned using a
microtome setting of 4 microns
• Measure Section Thickness – Interferometry
• Stain– H&E
• Measure intensity– Whole slide imaging
Measuring tissue thickness using vertical scanning interferometry
source: cnx.org
Tissue thickness using interferometric measurements
• Glass vs. paraffin
• Tissue was not measured
•Interferometer limitation
•Glass level variability
• Measurements taken at 6 locations repeatedly
How well are we using the interferometer?
Source Standard deviation
% Contribution
Total measurement (gage)
0.29 0.80%
Repeatability – equipment variation
0.29 0.79%
Reproducibility – operator variation
0.03 0.01%
Slide variation 3.20 99.20%
Total variation
3.21 100.00%
Equipment Variation
Operator Variation
Tissue ThicknessVariation
How good is our tissue thickness measuring system? - gage R & R
Equipment
variation – 0.79%
Operator
variation – 0.01%
Sample
variation – 99.20%
Slice thickness variation – by histologistHistologi
stNumbe
r of slides
Measured thickness average ± S.D. (m)
Combined
128 4.74 ± 0.16
1 64 4.65 ± 0.10
2 64 4.84 ± 0.16
• Nominal setting was 4 microns
• Both Histologists cut significantly thicker than 4 microns
• Both Histologists cut at significantly different thicknesses from each other
Manual vs. automated microtomy impact on tissue thickness
Histologist
Microtomesetting
Measuredthickness ± S.D. (m)
1 Automated 4.65 ± 0.13
Manual 4.65 ± 0.08
2 Automated 4.91 ± 0.16
Manual 4.76 ± 0.12
• Histologist 1 mean thickness was not impacted by microtome setting
• Both histologists had statistically significant more variability using the
automated setting as compared to the manual setting
Block influences tissue thickness
Tissue block
Measured thickness average ± S.D. (um)
Tissue one(n=32)
4.65 ± 0.13
Tissue two(n=16)
4.60 ± 0.12
Tissue three(n=16)
4.36 ± 0.12
• Histologist 1 was the cutter
• Automated setting used
• Tissue 3 was cut significantly thinner than tissues 1 & 2
Summary of tissue thickness measurement results
1. Histology (location within block, slice selection, soaking, etc.)• Difference in mean tissue thickness
2. Microtome setting – automated vs. manual• Both histologists were impacted by setting
3. Block• Blocks 1 and 2 were cut more thickly than block 3
Talk overview
• Introduction
• Tissue thickness variation– Using best histological practices
• Stain intensity variation due to tissue thickness
• The difference matters– Could impact algorithm functionality
Stain intensity variation due to tissue thickness - normal breast lymph node study
3 micron 4 micron
Objective – measure tissue thickness impacton stain intensity
• Tissue was sectioned and measured for thickness
• All slides were stained using the same method
• All slides were scanned using whole slide imaging and their average intensities were measured
Lymph node – 1 micron makes a measurable difference
Effects of tissue thickness on intensity
0
50
100
150
200
250
2 2.5 3 3.5 4 4.5 5
Tissue thickness (m)
Inte
nsi
ty
Intensity
Linear Fit
Talk overview
• Introduction
• Tissue thickness variation– Using best histological practices
• Stain intensity variation due to tissue thickness
• The difference matters– Could impact algorithm functionality
255 204 153 102 51 00
1000
2000
3000
4000
5000
Intensity level
Effects of tissue thickness on binning
2.62 micron
3.32 micron
3.43 micron
4.37 micron
Grey scale intensity differences
Pixel count
(N)
Summary
• Expected vs. measured is different
• The difference is quantifiable – Tissue thickness – Stain intensity
• The difference matters– Could impact algorithm functionality
• Tissue thickness and stain intensity correlate as expected
Further studies
• Intensity vs. tissue type
• Microtome bounce
• Histology vs. – Drift– Knife– Location in block– Degrees of fixation
Acknowledgments
Cindy Connolly
Wendy Lange
Allison Cicchini
Heather Free
Aaron Ewoniuk
Jonathan Hall
Mike Cohen, Ph.D.
David Clark, Ph.D.