Computational analysis of lipid droplet content

For every image, a mask corresponding to regions of dead or incompletely imaged cells was manually created. Each z-stack image was processed separately with a recursive thresholding method. The initial step was to separate cells from background using quantization to five levels. Further processing was done on the layer with the highest intensity by segmentation (Dougherty 2009). Dots with the lowest intensity were detected by this step. All segmented objects with area larger than 800 pixels or with non-circular shape were recursively taken to the next step of the segmentation process with a higher threshold value. The result of each level of the recursive thresholding method was a binary mask with segmented dots. The second step was to merge all segmented areas from the analyzed z-stacks. The last step was to remove detected objects that belonged to cells classified as dead or not present completely in the image, and objects having area smaller than 4 pixels. The final output of this process was a list of detected objects (i.e. lipid droplets) with extracted features. All processing was done in MATLAB Version: 9.2.0.556344 (R2017a) using Image Processing Toolbox Version 10.0 and Parallel Computing Toolbox Version 6.10.