Effect of True Area Sample Size on Variability in Endothelial Cell Density

(ECD) Results

Jackie Hai and Vivian XueHAI Laboratories, Inc.

Purpose

Evaluate the effect of sample area size on ECD measurement results and variability

Determine a recommended minimum sample size by comparing the results of cell counts performed within larger areas vs smaller areas

Sample Selection Criteria True area required for all samples

Images captured at high resolution

Low instances of pathology or abnormality

Random selection with variety of ECD, polymegathism, pleomorphism within normal range

Untrue Area: excessive concave curvature (circled)

Untrue Area: incorrect focal depth (cells out of focus)

True Area: flat, focused image with clear cell borderlines

True Area: flat, focused image with clear cell borderlines

Low Resolution Thumbnails(160x120 pixels)

Sample Area: 39981.10μm2

Cells Counted: 113Density: 2826

High Resolution Analysis(640x480 pixels)

Sample Area: 44843.25μm2

Cells Counted: 113Density: 2520

High Resolution Analysis(640x480 pixels)

Sample Selection Criteria True area required for all samples

Images captured at high resolution

Low instances of pathology or abnormality

Random selection with variety of ECD, polymegathism, pleomorphism within normal range

Counting Method

Variable frame counting method

Single-field (40,000μm2 – 60,000μm2)

Multiple-field (20,000μm2 – 25,000μm2)

Single-field variable frame count (60,000μm2)

HAI CAS / EB 2.10

Multiple-field variable frame count 1 of 3 (23,000μm2)

HAI CAS / EB 2.10

Multiple-field variable frame count 2 of 3 (22,000μm2)

HAI CAS / EB 2.10

Multiple-field variable frame count 3 of 3 (22,500μm2)

HAI CAS / EB 2.10

Results

ECD ranged from 1700 – 3500 cells/mm2

Result groupings: Class A (large samples): 40,000μm2 - 70,000μm2 Class B (small samples): 20,000μm2 - 25,000μm2

Standard Deviation: Class A (large samples): 6.27-63.11 (mean 26.3) Class B (small samples): 17.22-155.32 (mean 75.0)

Results

Standard Error (S.D. / √N): Class A (large samples): 3.14-31.56 (mean 13.2) Class B (small samples): 9.94-89.67 (mean 43.3)

Precision (S.E. / Mean): Class A (large samples): 0.0052 Class B (small samples): 0.0156

Results

In every instance: Class A Class B(40,000–70,000μm2) < (20,000–25,000μm2) Mean ECD Mean ECD

RPD ranged from 3.7% – 13.2%

Mean RPD of 7.9%

Conclusion

Larger sample sizes (>40,000μm2) yielded lower standard error and more precise cell count results

Cell counts performed in smaller sample sizes produce significantly higher ECD than larger sample sizes

Discussion

Rate of Change 1,000,000__ in Cell Density = Sample Area(cells per mm2) (in μm2)

Discussion

Sample size was based on area

Samples based solely on # of cells yield smaller sample size in high-density cornea

In smaller samples, ECD increases at a higher rate for every extra cell counted

Discussion

Smaller sample sizes compound possibility for increased margin of error Multiple small samples have longer total

perimeter than a single large sample

Human error factors Image quality Cell identification Tracing of cell boundaries

Recommendations1. Capture images at high resolution and

maximum screen size (640 x 480 pixels).

2. Select the largest sample area possible on the screen for cell counting.

3. If there are folds or distortions, select the largest contiguous true area possible.

4. For high resolution images free of distortions, a minimum selection of 50,000μm2 should always be possible.