Semi-automatic FISH quantification on digital slides

Two TMA samples with 20-19 representative HER2 cases were used in this study. Samples were selected from the archive of the 1st Department of Pathology and Experimental Cancer Research of the Semmelweis University, Budapest, Hungary. The survey was performed with the permission of the Institutional Review Board and Regional Ethics Committee of the Semmelweis University (RKEB) (permit no. 7/2006). Sample selection was based on the original IHC HER2 (c-erbB-2 / HER-2 / neu Ab-12, Thermo Fisher Scientific Inc. USA) scoring in order to use cases that cover all positivity ranges. Thus 24 HER2 positive and 16 HER2 negative cases were included in the TMAs. The formalin-fixed paraffin embedded specimens were stained using HER2 FISH pharmDx™ kit (K5331, Dako, Denmark). Scanned was performed using Pannoramic 250 FLASH (3DHISTECH Ltd., Hungary), that utilizes Plan-Apochromat 40x magnification, 0.95 numerical aperture objectives (Zeiss, Germany) and a PCO.edge 5.5 megapixel, scientific Complementary Metal-Oxide Semiconductor camera (Kelheim, Germany) for fluorescent image acquisition. One part of the image optimalization processes was integrated in the Pannoramic 250 FLASH scanner, which can set up the white balance and can make a shedding correction and a special fluorescent compensation as well. By applying the extended focus scanning method, the different recorded focus lanes (Z=5, step size=0,8 µm) were represented in one summarized virtual lane, which formed the basis of the image processing. The three fluorescent channels were recorded separately and represented the inputs of our FISH detection algorithm.

We used Visual C++ for developing the image analysis software application. The IPP (Intel Performance Primitives) library was used for image processing.

After the whole slide scanning, all TMA cores were scored by an expert pathologist, using Pannoramic Viewer 1.15.2 ver. software (3DHISTECH Ltd., Hungary). Maximum 40 cell nuclei were quantified, as detailed in the HER2 FISH pharmDx™ kit scoring guideline.

Comparisons were made between the semi-automated assessments, the scorings performed using the software platform and the original ICH scores.

As the data obtained show a non-normal distribution, non-parametric tests (Cohen’s kappa and Spearman rank-correlation) had to be applied, using the MedCalc for Windows v. 11.2.1.0 (MedCalc Software, Mariakerke, Belgium) software application. The strength of agreement was interpreted as proposed by Landis & Koch [7].

First we analyzed the blue (DAPI) channel, which represented all cell nuclei. The intensity dynamics and the geometrical morphology of the epithelial cell nuclei were investigated (Figure1A). After a locally adaptive contrast enhancement the intensity profile of the nuclei followed smooth shape, where the most frequent pixel intensities were from the central chromatin of the cell nuclei. A specific slope graph could be defined at the border of the nuclei in the region of the marginal chromatin and the nuclear membrane. Using these findings, the intensity and the geometrical features of an ideal cell nucleus could be described. This “ideal nucleus” was used hereafter as a nuclei filter in the detection algorithm.

After extended focus recording, all FISH spots were represented in focus plane. Thus the geometrical and intensity parameters of the signals were similar in one channel (Figure 1B, 1C, 1D). This allowed us to apply a general spot detection algorithm for the whole slide. Our algorithm used a local intensity amplifying mechanism, followed by a threshold process, the latter dependent on the quality of the digital slide.

The actual spot numbers were identified for each of the detected cell nucleus and based on this the cell nuclei were classified using the usual Her2 classification rule. Three groups were defined; normal (red signal and green signal were represented equally in the cell nucleus), amplification (the ratio of the red signal and green signal was higher than one per nucleus) and artifact (others).

A Microscopical Image Segmentation Profile (.misp file) was defined based on the morphological parameters of the nuclei (i.e. radius, area and circularity), the contrast and the intensity of the slide. This profile was used to make investigation on 39 TMA spots. The results of the measurement were saved into an Extensible Markup Language (XML) file and were compared with the results of the Her2 immunostain.

Results of the TMA cores (Figure 2) could be used in the statistical analysis, in diverse extents. The number of nuclei identified and scored using the software platform within each TMA core higly exceeded (mean 279.84 cells, min. 26-max. 939 cells) the number of cells suitable for enumeration of borderline cases (i.e 40 cells) as defines in the FISH pharmDx™ kit scoring guideline. Therefore the ratio of normal and amplified cells has been standardized to 40 cells/TMA core prior to conducting the statistical analysis.

An almost perfect agreement has been found between the semi-automated scoring and the results provided by the software platform (Table 1.). Apposite to this result, both the semi-automated investigation and the automated scoring showed only substantial agreements with the initial ICH HER2 score.