A) To provide an automated, unbiased approach for estimating vessel densities, we wrote and validated a custom macro in FIJI (
Schindelin et al., 2012). In vivo imaging stacks of dye filled cortical vessels (see maximum z-projection in top left and side view projection of image stack in top right) were manually split into sub-stacks of 10 images (2 µm z-steps between images). Sub stacks were then each maximally projected in the z plane, and each automatically thresholded using the built in ImageJ threshold function Triangle (
Zack et al., 1977), which was determined to best capture vascular signal through trial and error. The area and fractional vascular volume of vascular signal (number of black pixels after thresholding) was measured for each sub-stack image projection after applying a median filter (radius of 2 pixels) to eliminate speckling. Thresholded sub-stacks were then binarized and skeletonized using built in functions (
Arganda-Carreras et al., 2010) and total vascular length was taken as the number of skeleton pixels. From the vascular length, average vessel width (w = A/L) was also calculated. (
B–C) Graphs show that both the fractional vascular volume (v/v) and the estimated number of capillaries per imaging stack (0.02 mm
3 volume) were sensitive to the volume of images projected. For fractional vascular volume, projecting 20 µm sub-stacks led to estimates of ~0.01% vascular volume which closely matches published data (
Blinder et al., 2013;
Schmid et al., 2017). As for capillary number, we validated 20 µm sub-stack z projections by comparing automated estimates with those derived from blinded manual counts (data from four mice, two imaging areas per animal). (
D) Box and whisker plots (+is mean) showing close agreement between manual and automated estimates of capillary number per imaging stack [paired t-test t
(3)=0.33, p=0.76]. Error bars are S.E.M.