Color Calibration on Whole Slide Imaging Scanners

Presented By: Vipul Baxi, Omnyx LLC Intersystem Color Variability

Pantanowitz, L. (2010). "Digital images and the future of digital pathology." Journal of Pathology Informatics 1(1): 15-15

Importance of Color Calibration

• Provide Standardization o Amongst digital scanners produced by the same manufacturer o Amongst digital scanners with diverse technology and scanning components o Consistency and Accuracy of CAD algorithms • Transition and adoption of digital pathology o Pathologists get the same view of the sample as they would under a microscope o Prevent possible misdiagnosis due to color inaccuracy What is Color Calibration?

Definition: Process to measure and/or adjust the color response of a device (input or output) to establish a known relationship to a standard color space • Widely used in the field of Photography o Macbeth Color Checker chart (24-patch) • Total Color Management

Imaging System Display

Digital Pathology Imaging System

3 Components affecting Color 1. Light Source 2. Optics 3. Sensor 2

http://www.richardwheeler.net/contentpages/image.php?gallery=Scientific_Illustration&img=Epifluorescence_Microscope&type=jpg The Color Space Chromaticity Plot • Spectral Locus o Represents the color gamut visible by humans o Visible range: 380nm to 700nm • sRGB Locus (triangle) o World standard for digital images, printing and the internet o Used in current LCD monitors, HDTVs Color Target Slide for Microscopy

Color Target Film Reference Values • NIST traceable measurements of color patches • Chromaticity plot of color patches

Chromaticity plot of Color Checker slide

Spectral Locus sRGB Locus Target Slide Assembly * Color Patches

Cover Slip Y

Color Target

Glass Slide X Calibration Procedure Color Calibration Matrix Image Acquisition Digital R1 G1 B1 Camera Transformation to Reference Values R2 G2 B2 Objective 3 3 3 Lens R G B Roffset Goffset Boffset

Convert the color patch data into CIE standard Target slide color space

1. Acquire image of each patch 1. Compare to Reference values 2. Calculate CIE Color Space values and obtain best transformation The transformation matrix can be applied to input color patches and objectively measure the color difference before and after Color Difference – CIE94

• Convert the color space into CIELAB coordinates

• The higher the ∆E*94, the larger the color difference • Useful in determining:

o How inaccurate is the color compared to the reference? o How close is the color after using the color correction matrix? Correction of Color Patches

• ∆E*94 before any correction is high and easily noticeable

• ∆E*94 reduces considerably after applying correction • Current industries (graphics and Input textile) aim to obtain a mean ∆E*94 below 5 for their processes Corrected

Correction of Color Patches

Input

Corrected

Mean ∆E*94 = 1.70

Max ∆E*94 = 4.85 Digital Image Vs. Glass

Manual Review

Digital Images Review

IHC H&E Special

Digital Review Score each image on the following scale:

5: Identical – There is no noticeable difference in color between the glass and digital image 4: Similar – The color is very close, with subtle differences in certain features 3: Noticeably Different – The color is noticeably different, but should not affect diagnosis 2: Significantly Different – The color is extremely different, and may possibly affect diagnosis 1: Misrepresentation – The color is completely wrong

Comparison to Glass

STAIN Uncorrected Vs. Glass Corrected Vs. Glass H&E 4 5 H&E 4 5 H&E 4 5 H&E 3 5 Uncorrected Av g Score H&E 4 4 = 3.8 ~Similar PIN4 4 5 NEGATIVE 4 5 Corrected Av g Score CK20/Ki67 4 5 = 4.9 ~Identical NEGATIVE 4 5 COLLOIDAL IRON 3 5 PAS 4 5 GMS 4 5 Average 3.83 4.92

Uncorrected Image vs. Glass • “Blue is faded on the WSI (not as washed out on the glass)” • “Pink/stromal color is faded”

Corrected Image vs. Glass • “Staining matches, no noticeable difference • “Blue and pink are right Overall Color Impression

Uncorrected Image vs. Glass • “The blue and the brown are little faded” • “If intensity affected scoring of the slide, the score might change”

Corrected Image vs. Glass • “Exactly like glass” Overall Color Impression

Uncorrected Image vs. Glass • Pink is good, but the WSI’s blue color is a little too purplish”

Corrected Image vs. Glass • “Good, cannot detect a difference” Value of Color Correction

• The spectral data measured on the custom color target shows that all reference colors lies within the sRGB space • The ∆E94 considerably reduces after applying color correction • Majority of the corrected whole slide images, regardless of stain, were rated identical to the color under the microscope

Next Steps • Develop a Pathology specific color target • Total Color Manangement References

1. Pantanowitz, L. (2010). "Digital images and the future of digital pathology." Journal of Pathology Informatics 1(1): 15-15 2. Al-Janabi, S., Huisman, A., et al. (2011). "Digital pathology: current status and future perspectives." Histopathology: 1-9 3. Thomsen, K. (2000). "A Euclidean color space in high agreement with the CIE94 color difference formula." Color Research & Application 25(1): 64-65 4. Koren, N. "Imatest - Color Correction Matrix." Imatest. Imatest, 2009. Web. 08 Sept. 2011. . 5. Lindstrom, P. (2008). "Delta E Blues: The Science of Color Perception." Seybold Report: Analyzing Publishing Technologies 8(3): 13 6. Yagi , Y., Gilbertson, J. R. (2005). "Digital Imaging in Pathology: the case for standardization." J Telemed Telecare 11(3): 109-116 7. Hubel, P. M., Finlayson, G. D., et al. (1997). "Matrix Calculations for Digital Photography." Fifth Color Imaging Conference: Color Science, Systems and Applications: 105-111 8. Yagi , Y. (2011). "Color Standardization and Optimization in Whole Slide Imaging." Diagnostic Pathology 6(Suppl 1): 1-15 Acknowledgements

• Dr. Michael C. Montalto, VP Clinical & Scientific Research, Omnyx LLC • Dr. Richard R. McKay, Principal Scientist, Omnyx LLC

Questions?