3. Purpose
Quantify the relationship between
focal distance and image distortion
in specular microscopy
Determine predictive model of cell
counting error based on amount of
deviance from true endothelial cell
shape and size
8. Method
Images captured 30μm, 40μm, 50μm,
30 40 50
60μm, 70μm, 80μm, 90μm and 100μm
60 70 80 90 100
from baseline focal distance
10 sample area measurements per
image (single blind trial)
Determine mean area of each image
Simple linear regression model
22. True Endothelial
Area Cell Density
42,061μm2 2377 cells/mm2
Focal Measured Endothelial
Deviance Area Cell Density
F=0μm 42,061μm2 2377 cells/mm2
F=30μm 39,893μm2 2507 cells/mm2
F=60μm 37,792μm2 2646 cells/mm2
F=100μm 36,711μm2 2724 cells/mm2
23. Qualitative Properties
In Focus Out of Focus
Cell borders are bold Cell borders are fuzzy
and clearly visible and poorly visible
High contrast between Low contrast between
cell borders and interiors cell borders and interiors
Cell surfaces are even Cell surfaces are uneven
and consistently lit and/or have hot spots
Cell morphology includes Cell morphology appears
regular polygons flattened or stretched
28. Discussion
Minimizing focal deviance is
essential to capturing true area
and attaining accurate cell density
Higher resolution images allow
better judgment of focal deviance
Lower resolution images obscure
focal deviance due to compression
33. Recommendations
Allow donor tissue to warm up to
room temperature
Use both coarse and fine Z-knobs
to attain optimum focus
If it is difficult to focus, change
the angle of the chamber or vial
34. Recommendations
Capture specular images at the
highest possible resolution (e.g.
640x480 pixels)
Select flat areas of endothelial
cells to perform density analysis
http://www.hailabs.com/specular-microscopy
Editor's Notes
Today I’m going to talk about the effect of specular focal distance on endothelial cell counting accuracy. This is a continuation of our specular microscopy series started last year, where as the manufacturer we’ve been asked to give recommended best practices based on scientific methodology.
We’d like to thank the North Carolina Eye Bank for being very helpful with providing research cornea for this study.
The purpose of our analysis was two-fold: first, to quantify the relationship between the focal distance of the microscope and any resulting image distortion, and secondly, to formulate a predictive model of cell counting error based on the amount of deviance from true size and shape of the cells.
It’s common knowledge that accurate cell counts depend on high quality images. We’ve also observed, anecdotally, that samples captured out of focus throw off the cell density when compared to
the same samples captured in focus. We wanted to take this observation a step further and actually describe the mathematic relationship between focus and cell density.
Here’s what we did. First, we took a high resolution specular image of a standard calibration lens and established a baseline pachymetry measurement of zero micrometers to represent an image that is totally in focus.
This is the baseline “zero” calibration image.
Then, on the same calibration lens, we captured a series of increasingly out of focus images, at 30, 40, 50, 60, 70, 80, 90 and 100 micrometers distance from the baseline. We used a single blind trial to take 10 sample area measurements for each image, for a total of 80 samples. Based on this data, we determined the mean area of a section of the calibration grid at each level of focal distance and plotted a simple linear regression line to model the behavior.
Here’s an example of the sampling method, performed on a focal deviance of 30 micrometers.
Here we are at 60 micrometers.
And 100 micrometers. Note the mean area of 8467 square micrometers for these four blocks is much smaller than the mean area of 10,000 square micrometers for the same four blocks in the baseline “zero” image. This is because as you go out of focus, the image becomes distorted and appears smaller. This shrinking effect carries over to endothelial cells, and I’ll show you examples later in the presentation.
Here are the results in graphical form, illustrating the relationship between focal deviance and measured area. The mean areas are in blue, and the “best fit” regression line is in red.
If you look at the numbers, you’ll notice an inverse relationship between focal deviance and area. The more out of focus you go, the lower the area becomes.
Using the linear regression model, we can extrapolate from focal deviance to area to cell density. This table shows the cell density error at each 10-micrometer increment of focal deviance. The green level represents a smaller error, a difference in density less than 100 cells per square millimeter. The yellow level represents a difference in density greater than 100 cells per square millimeter. And the orange level represents the largest error, a difference in density greater than 200 cells per square millimeter.
Let’s say you have a cornea with an ECD of 2500. That’s 2500 at zero focal deviance, totally in focus. The same cornea at 30 micrometers focal deviance now has a density reading of over 2600. At 60 micrometers, 2750. And at 100 micrometers out of focus, the density reads over 2950.
The effect is even more pronounced at higher densities. For a cornea with an ECD of 3000, the difference in density from 0 to 100 micrometers of focus is over 500 cells per square millimeter.
We can draw the conclusion that the cell density error is negligible for a focal deviance of 0 to 29 micrometers, non-negligible from 30 to 59 micrometers, and problematic from 60 to 100 micrometers and above. That’s enough math, let’s look at a real-life example.
Here’s a specular image of a donor cornea captured 100 micrometers out of focus.
Here’s the same cornea, with the same group of cells selected, at 60 micrometers out of focus.