After preprocessing, we segmented all major tissue components. Images were converted into binary images by applying histogram-based threshold. For this purpose, the mode and standard deviation (σ) were determined for each channel of an image stack. Threshold levels were set to mode+σ for the WGA channel and mode+2σ for the DAPI, vimentin and α-SMA channels. We used a lower weighting factor for σ to threshold WGA channels because the high ratio of voxels with WGA signal (≥ 20%) led to a relatively higher standard deviation versus other channels. Images were then filtered with a binary median filter (radius 1). The segmented WGA signal was used to segment each cardiomyocyte by means of a watershed-based method as described previously.34 This method created labeled segments for each volume enclosed by WGA signal. At first, we identified myocytes. Subsequently, vessel lumens were identified within the non-myocyte segments. Vessel lumens were then dilated with a radius of 1 μm to include the vessel walls and endothelium. Possible overlap with the myocyte space was subtracted after the dilation. Cardiomyocyte and vessel volumes were then subtracted from the segmented WGA, DAPI, vimentin and α-SMA volumes. The remaining DAPI was identified as non-cardiomyocyte nuclei and merged with the remaining vimentin and α-SMA volumes. Using distance maps, we classified voxels of the merged volume closer to vimentin than to α-SMA as fibroblasts, while other voxels were classified as myofibroblasts. Finally, remaining non-classified segments from the watershed algorithm were merged with the remaining WGA volume and classified as extracellular space.
We determined volume ratios of all tissue components. Also, we calculated the number of neighboring myocytes by selecting complete cardiomyocytes, which were not truncated at image borders. We generated distance maps from the complete myocytes and then identified and counted all myocytes within a distance of 0.5 μm to the outer sarcolemma.
For 2D visualizations of images we used customized software. For 3D visualizations we applied Paraview (Kitware, Clifton Park, New York, USA), VolView (Kitware) and customized software.
We determined volume ratios of all tissue components. Also, we calculated the number of neighboring myocytes by selecting complete cardiomyocytes, which were not truncated at image borders. We generated distance maps from the complete myocytes and then identified and counted all myocytes within a distance of 0.5 μm to the outer sarcolemma.
For 2D visualizations of images we used customized software. For 3D visualizations we applied Paraview (Kitware, Clifton Park, New York, USA), VolView (Kitware) and customized software.
