4.4. Image Analysis

TMA slides were scanned using an Aperio scanner, with a 40× objective. The high-resolution images were analyzed using QuPath [27] version 0.1.2. The workflow consisted of (i) color deconvolution, (ii) identifying the CMA or TMA cores, (iii) segmenting the tissue region in each core, (iv) isolating nuclear and cytoplasmic regions of interest, (v) estimating the abundance of the deconvolved red component in each cell, and finally (vi) calculating the percentage of positive cells in each core. Three-color deconvolution was performed using vectors calibrated visually on the image data, following the procedure outlined in the software documentation. All images were deconvolved using the same stain vectors. Steps (iii) to (v) were performed using the default algorithms in QuPath. For step (vi), each cell was considered positive if the red component intensity was above a threshold, calculated independently for each CMA or TMA slide as the average between the 5th and 95th percentile of red intensities across the slide. All algorithms and parameters for the analysis in QuPath were recorded in a script for repeatability. Color composition of blue and red are components of the color-deconvolved image. The original image was deconvolved using the “Color Deconvolution” plugin in Fiji (download data May 5th 2020) with the “FastRed FastBlue DAB” color settings, producing a blue, a red, and a brown image. The blue and red images were converted to RGB and merged using the “Image Calculator” with “Min” settings. No contrast adjustments were made.