Automated IHC Image Analysis Plugin

The IHC images used were stained with DAB and hematoxylin. The result of color deconvolution leads to the production of three images, namely, DAB, hematoxylin and a complimentary image. In a previous study, we reported development of an ImageJ compatible plugin for analyzing cytoplasmic staining pattern by assigning a histogram profile for the deconvoluted DAB image [21] . Now, within the scope of the current plugin development, we envisioned automating the whole process by integrating deconvolution, histogram profiling and scoring by a simple choice of the program menu. Additionally, it integrates methods with a wider scope of analyzing various marker proteins displaying cytoplasmic or nuclear staining patterns (Table 1).
In digital image analysis, the pixel intensity values for any color range from 0 to 255, wherein, 0 represents the darkest shade of the color and 255 represent the lightest shade of the color as standard. A total of 1703 images were analyzed independently with the help of two expert pathologists and were assigned a score as high positive (3+), positive (2+), low positive (1+) and negative (0). In the current method development, the next step was assigning a histogram profile which is a plot between the intensity values of the pixels (X axis) vs. the number of pixels representing the intensity (Y axis). Keeping in view the standard grading procedure, the histogram profile was divided into 4 zones, viz. high positive, positive, low positive and negative. These four zones were equally divided on the pixel color intensity bar (as indicated in Figure S1A).
The zones were visually identified by using the threshold feature of Image menu of the ImageJ program [22] (link). To begin with, the intensity values were grouped into bands of 10 and the corresponding regions in the image were confirmed by using the threshold feature. Thus initially all the intensities from 0 to 10 were turned red on an image with a known pathological score of high positive. Then, in addition to the previous band, the next band, from 11 to 20 were turned red and the same was continued until all the pixels of the brown shades were assigned a threshold and the range for the high positive zone was determined. Similarly, zone containing the lightest color shade of pixel intensities was also determined using an image with a known pathological score of negative. This was because once the highest intensity (high positive) and the least intensity (negative) zones were determined, it would help towards the better determination of the size of the intermediate (positive and low positive) zones. It was found that the region between 0 and 60 contained pixels of the high positive stained images. Similar was noted on samples with known pathological lower scores to optimize the correct range. The process was repeated for at least 70 images of same intensity of the color shade. The intensity range for the positive zone was found to be ranging from 61 to 120; 121 to 180 for the low positive zone; and 181 to 235 for the negative zone, respectively (Figure S1). It was determined that the pixels with intensity values ranging from 235–255 predominantly represent fatty tissues which are occasionally present but do not typically contribute to pathological scoring and were therefore excluded from the score determination zones. The intensity ranges determined visually were confirmed by plotting a histogram using Microsoft Excel (data not shown).