A robust automated measure of average antibody staining in immunohistochemistry images - PubMed (original) (raw)

A robust automated measure of average antibody staining in immunohistochemistry images

Kingshuk Roy Choudhury et al. J Histochem Cytochem. 2010 Feb.

Abstract

Identifying and scoring cancer markers plays a key role in oncology, helping to characterize the tumor and predict the clinical course of the disease. The current method for scoring immunohistochemistry (IHC) slides is labor intensive and has inherent issues of quantitation. Although multiple attempts have been made to automate IHC scoring in the past decade, a major limitation in these efforts has been the setting of the threshold for positive staining. In this report, we propose the use of an averaged threshold measure (ATM) score that allows for automatic threshold setting. The ATM is a single multiplicative measure that includes both the proportion and intensity scores. It can be readily automated to allow for large-scale processing, and it is applicable in situations in which individual cells are hard to distinguish. The ATM scoring method was validated by applying it to simulated images, to a sequence of images from the same tumor, and to tumors from different patient biopsies that showed a broad range of staining patterns. Comparison between the ATM score and manual scoring by an expert pathologist showed that both methods resulted in essentially identical scores when applied to these patient biopsies. This manuscript contains online supplemental material at http://www.jhc.org. Please visit this article online to view these materials.

PubMed Disclaimer

Figures

Figure 1

Figure 1

Two series of cartoons depicting the methodology for calculation of the Allred score. The green color identifies unstained cells, whereas the gray, dark gray, and black colors identify cells stained to different intensities. (A) Series in which the stain intensity is constant (at maximum), and the proportion of stained cells increases from left to right. (B) Series in which the proportion of stained cells is constant (at 1/3), and the stain intensity increases from left to right (from none to maximum). With permission, Allred (2008).

http://www.asbd.org/images/D3S9%20-%20Craig%20Allred.pdf

Figure 2

Figure 2

(A) Image of a typical immunohistochemistry (IHC)-stained slide. Background staining is generally a light blue with this technique, and represents cells that do not express the antigen of interest. The brown stain is the product of a colorimetric assay that is of standard use in clinical IHC measurements, and represents cells that are positive for the antigen of interest. (B) Image representing the intensity of brown staining for the slide in A. Brighter pixels indicate stronger staining, dark pixels indicate no staining. (C) Thresholded version of the intensity image: pixels with intensity above 100 are identified as white; the remainder are black. (D) Another thresholded version of the intensity image: pixels with intensity above 70 are identified as white; the remainder are black. (E) Curve showing the percent area stained as a function of the threshold. Bar = 200 μm.

Figure 3

Figure 3

Sequence of hypoxia inducible factor 1 (Hif1) images. Randomly chosen fields from a single tissue biopsy slice, stained for Hif1 as described in the Materials and Methods section. Viewing fields were blindly selected, and images taken with an attached digital camera. Fields of view that included no neoplastic tissue were excluded from further analysis. Bar = 200 μm.

Figure 4

Figure 4

Images of Hif1 IHC from both head and neck cancer and cervical cancer tissue slices. These represent a spectrum of staining, from no stain or very low stain (A,J) to different representations of high stain (C,E,G) and intermediate staining between the two extremes (B,D,F,H,I). The images were specifically chosen to represent a broad variation in staining pattern. Bar = 200 μm.

Figure 5

Figure 5

Images of vascular endothelial growth factor (VEGF) IHC from the same cancer types. These represent a spectrum of staining, from no stain and very low stain (B,C) to different representations of high stain (A,I,J), with intermediate staining patterns in the others (D–H). The images were specifically chosen to represent a broad variation in staining pattern. Bar = 200 μm.

Figure 6

Figure 6

Comparison of manual and automated scores. (A,B) The diagonal line is y = x; manual scores are on the x-axis, automated scores are on the y-axis. (A) Comparison of percent stained scores. (B) Comparison of overall scores. (C) Effect of histological heterogeneity on comparison of manual and automated scores.

References

    1. Allred DC (2008) Estrogen and progesterone receptors as predictive markers for breast cancer and DCIS. http://www.asbd.org/images/D3S9%20-%20Craig%20Allred.pdf (accessed November 9, 2009)
    1. Allred DC, Harvey JM, Berardo M, Clark GM (1998) Prognostic and predictive factors in breast cancer by immunohistochemical analysis. Mod Pathol 11:155–168 - PubMed
    1. Altman DG, Lausen B, Sauerbrei W, Schumacher M (1994) Dangers of using “optimal” cutpoints in the evaluation of prognostic factors. J Natl Cancer Inst 86:829–835 - PubMed
    1. American Society of Clinical Oncology (1998) Update of recommendations for the use of tumor markers in breast and colorectal cancer. J Clin Oncol 16:793–795 - PubMed
    1. Grandis JR (2006) Prognostic biomarkers in head and neck cancer. Clin Cancer Res 12:5005–5006 - PubMed

Publication types

MeSH terms

Substances

LinkOut - more resources