Volume 8 Supplement 1
A multistep image analysis method to increase automated identification efficiency in immunohistochemical nuclear markers with a high background level
© Lejeune et al; licensee BioMed Central Ltd. 2013
Published: 30 September 2013
In anatomical and surgical pathology, the customary method of manual observation and measurement of immunohistochemically stained markers from microscopic images is tedious, expensive and time consuming. There is great demand for automated procedures for analyzing digital images (DIs) of these markers  given that they reduce human variability in the evaluation of stained markers [2, 3] and increase the speed and efficiency of the analysis . Computerized DI analysis software generally involves a stained objects/nuclei segmentation method to detect and quantify the number of positively stained markers in combination with the standard evaluation of their morphometric and/or densitometric features [5, 6]. However, automatic segmentation often fails due to the presence of spurious stain deposits in tissue sections (background). The “removal” of the background from noisy DIs, so that only the objects of interest are identified, is difficult due to the color values of pixels in the nuclei and background overlapping during the color segmentation processes.
We previously developed an automated macro that allows quantification of several nuclear markers in various neoplasic tissues . In an attempt to standardize the immunohistochemical analysis and to improve cell detection, we propose a new procedure that quantifies only positively stained nuclei even though they have a similar color to that of the surrounding tissue. The aim of this work was to develop a single automated procedure that allows images to be analyzed irrespective of whether the spurious stain deposit in background is present or absent. The multistep process includes algorithms that permit this discrimination so that the appropriate procedure for optimal quantification can then be applied.
Materials and methods
Histological sections of lymphomas and breast cancer tissues, previously immunohistochemically stained with standardized protocols [8, 9], were selected from the archives of the Department of Pathology of the Hospital de Tortosa Verge de la Cinta, Catalonia, Spain. Staining was performed with monoclonal antibodies directed against the nuclear protein estrogen receptors (ERs; clone NCL-ER-6F11, Novocastra, Newcastle upon Tyne, UK), progesterone receptors (PRs; NCL-PGR-312, Novocastra), Ki-67 (clone MIB-1, Dako, Carpinteria, CA, USA) and FOXP3 (clone FOXP3-236A/E7, CNIO, Spain). The entire process was standardized to ensure high reproducibility and brown staining homogeneity, which are very important requirements for image analysis . This study received institutional review board approval.
Procedure developed in the new procedure
Quantification and statistical analysis
The mean of two manual counts made by two trained observers was taken as the reference value (gold standard) to validate the results obtained with the old macro and the newly proposed procedure. The comparisons made were: manual 1 versus manual 2 readings; mean of manual readings versus old macro reading; mean of manual readings versus new procedure reading; and the old macro versus the new procedure readings. The extents of agreement between the manual and automatic results were evaluated with Bland-Altman and Kaplan-Meier analyses with their corresponding graphs. Bland-Altman graphs illustrate the differences between the compared methods with respect to the mean of each paired count. Kaplan-Meier curves portray the conditional probability of observing differences between results obtained from the methods compared. All statistical analyses were carried out with SPSS 19.0.
Results and discussion
As previously demonstrated, image complexity relative to the number of positively stained nuclei may affect the automated nuclear quantification . In the present study, DIs were divided as before into a low-complexity group (≤100 positively stained nuclei/images) and a high-complexity group (>100 positively stained nuclei/image). As observed in Figure 5B2, the small differences in the counts between the manual method vs. the old macro and between manual counts vs. the new procedure were similar in low-complexity images. However, in high-complexity images (Figure 5B1), larger count differences were observed, although those between the manual counts vs. the new procedure were much lower than those between the manual counts vs. the old macro.
A general point regarding the proposed method is that, for high-complexity DIs with background (Figure 6A3), the quantification of nuclear markers obtained with the new procedure (green curve) was closer to the gold standard (manual method, black curve) than with the old macro (red curve). For high-complexity DIs without background (Figure 6A4), the two automated methods gave the same results (red and green curves). The presence or absence of background did not appear have a great influence on the quantification of nuclear markers in low-complexity DIs (Figure 6A1-A2), probably because these images have a lower background level than the other images.
Despite specific and careful preparation of tissues and the use of blocking buffers, strong background staining can sometimes mask the detection of the target antigen during automated analysis. The results of the method presented in this paper are promising since the selective identification of brown color ranges and morphological parameters of selected objects in DIs enables the background to be discriminated during the automated localization and quantification of specific stained nuclei. The principle of this approach is applicable to all quantitative nuclear signals and should prove useful in a variety of tumor specimens, irrespective of the immunohistochemical techniques employed.
ML, CL and JJ conceived the study. CL designed and implemented the algorithms. JB and AK provided IT support. RB and JJ carried out the manual quantification, discussed the methods and modified them. VG, BT and CC realized the automated quantification with the two automated methods and designed the database. AR analyzed the data. ML, CL and AK drafted the manuscript. All authors revised the manuscript and approved the final version.
List of abbreviations used
forkhead box protein 3
Tagged Image File Format
- Rojo MG, Bueno G, Slodkowska J: Review of imaging solutions for integrated quantitative immunohistochemistry in the Pathology daily practice. Folia Histochem Cytobiol. 2009, 47 (3): 349-354.PubMedGoogle Scholar
- Gavrielides MA, Gallas BD, Lenz P, Badano A, Hewitt SM: Observer variability in the interpretation of HER2/neu immunohistochemical expression with unaided and computer-aided digital microscopy. Arch Pathol Lab Med. 2011, 135 (2): 233-242.PubMedGoogle Scholar
- Lejeune M, Jaén J, Pons L, López C, Salvadó MT, Bosch R, García M, Escrivà P, Baucells J, Cugat X, Álvaro T: Quantification of diverse subcellular immunohistochemical markers with clinicobiological relevancies: validation of a new computer-assisted image analysis procedure. J Anat. 2008, 212 (6): 868-878. 10.1111/j.1469-7580.2008.00910.x.PubMed CentralView ArticlePubMedGoogle Scholar
- Bolton KL, Garcia-Closas M, Pfeiffer RM, Duggan MA, Howat WJ, Hewitt WJ, Yang XR, Cornelison R, Anzick SL, Meltzer P, Davis S, Lenz P, Fiqueroa JD, Pharoah PD, Shermand ME: Assessment of automated image analysis of breast cancer tissue microarrays for epidemiologic studies. Cancer Epidemiol Biomarkers Prev. 2010, 19 (4): 992-999. 10.1158/1055-9965.EPI-09-1023.PubMed CentralView ArticlePubMedGoogle Scholar
- Markiewicz T, Wisniewski P, Osowski S, Patera J, Kozlowski W, Koktysz R: Comparative analysis of methods for accurate recognition of cells through nuclei staining of Ki-67 in neuroblastoma and estrogen/progesterone status staining in breast cancer. Anal Quant Cytol Histol. 2009, 31 (1): 49-62.PubMedGoogle Scholar
- Di Cataldo S, Ficarra E, Acquaviva A, Macii E: Automated segmentation of tissue images for computerized IHC analysis. Comput Methods Programs in Biomed. 2010, 100 (1): 1-15. 10.1016/j.cmpb.2010.02.002.View ArticleGoogle Scholar
- Lopez C, Lejeune M, Salvado MT, Escrivà P, Bosch R, Pons LE, Alvaro T, Roig J, Cugat X, Baucells J, Jaén J: Automated quantification of nuclear immunohistochemical markers with different complexity. Histochem Cell Biol. 2008, 129 (3): 379-387. 10.1007/s00418-007-0368-5.View ArticlePubMedGoogle Scholar
- Lopez C, Lejeune M, Escriva P, Escrivà P, Bosch R, Salvadó MT, Pons LE, Baucells J, Cugat X, Alvaro T, Jaén J: Effects of image compression on automatic count of immunohistochemically stained nuclei in digital images. J Am Med Inform Assoc. 2008, 15 (6): 794-798. 10.1197/jamia.M2747.PubMed CentralView ArticlePubMedGoogle Scholar
- Alvaro-Naranjo T, Lejeune M, Salvado-Usach MT, Bosch-Príncep R, Reverter-Branchat G, Jaén-Martínez J, Pons-Ferré LE: Tumor-infiltrating cells as a prognostic factor in Hodgkin's lymphoma: a quantitative tissue microarray study in a large retrospective cohort of 267 patients. Leuk Lymphoma. 2005, 46 (11): 1581-1591. 10.1080/10428190500220654.View ArticlePubMedGoogle Scholar
- Seidal T, Balaton AJ, Battifora H: Interpretation and quantification of immunostains. Am J Surg Pathol. 2001, 25 (9): 1204-1207. 10.1097/00000478-200109000-00013.View ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.