Streamlining volumetric multi-channel image cytometry using hue-saturation-brightness-based surface creation

Image cytometry is the process of converting image data to flow cytometry-style plots, and it usually requires computer-aided surface creation to extract out statistics for cells or structures. One way of dealing with structures stained with multiple markers in three-dimensional images, is carrying out multiple rounds of channel co-localization and image masking before surface creation, which is cumbersome and laborious. We propose the application of the hue-saturation-brightness color space to streamline this process, which produces complete surfaces, and allows the user to have a global view of the data before flexibly defining cell subsets. Spectral compensation can also be performed after surface creation to accurately resolve different signals. We demonstrate the utility of this workflow in static and dynamic imaging datasets of a needlestick injury on the mouse ear, and we believe this scalable and intuitive approach will improve the ease of performing histocytometry on biological samples.


Supplementary Figure 2 Screenshot showing cell subsetting using statistics filters in Imaris.
One way of defining cell subsets is to use the statistics filters in Imaris. Zoomed in view of tracks selected using track intensity median hue values between 2.13 x 10 4 and 2.93 x 10 4 . Selected cells tracks are highlighted in yellow.

Supplementary Figure 3 Spectral compensation using traditional surface creation and HSB surface creation workflows.
(a) Spectral compensation using traditional surface creation. Images are acquired, and the raw images are corrected based on the spillover coefficients derived from the single stains. Cell subsets are then visualized during the creation of co-localization channels, followed by surface creation. (b) Spectral compensation using HSB surface creation. Images are acquired, and the information from all the channels merged into the brightness channel. Surfaces are then created based on the brightness channel, and the statistics extracted. Spectral compensation can then be carried out based on the spillover coefficients derived from the single stains. The cell subsets can then be visualized through analysis.
Supplementary Figure 4 Application of HSB surface creation to more than three channels. This example of a spleen section stained with 6 markers illustrates how HSB surface creation can be used for an image with more than 3 channels. The channels can be combined into a maximum intensity channel for surface creation, and specific channel combinations can be used for the hue-saturation-brightness conversion. Created surfaces are then rendered in the hue channel to check that the segmentation is accurate. The original channels can then be loaded for statistics extraction from the cell surfaces which can then be exported for histocytometry analysis or objective clustering using tSNE or k-means clustering. For the purposes of more accurate segmentation, we recommend that users include at least one cytoplasmic marker in their panel. Experiment was carried out once. For an image C with the number of cellular markers n, where r is the number of markers that are multiplexed C(n,r), the number of possible cell subsets is expressed by the mathematical formula : n C 1 + n C 2 + n C 3 +….+ n C r-1 + n C r or 2 n -1.  Table 2 Precision & recall rates for dermal dendritic cells (dDCs) and neutrophils (Neus) after filtering by track median hue intensity.

Supplementary
Precision & recall rates for two representative time-lapse imaging experiments of dDCs and Neus responding to a needlestick injury were calculated and averaged. Precision was calculated in the following manner: true positives/ (true positives + false positives). Recall was calculated in the following manner: true positives/ (true positives + false negatives).

Supplementary Note 2 Equations for conversion of image from Red-Green-Blue (RGB) colour model to Hue-Saturation-Brightness surface creation workflow
The Red-Green-Blue values are scaled by the dynamic ranges set by the user (i.e. maximum and minimum values are set by the user) to change the range to (0, 1) for each of the channels.
, where M represents the maximum RGB value of the image.
, where m represents the minimum RGB value of the image. The saturation calculation is inverted in the HSB surface creation workflow (as compared to traditional HSB models) to ensure that the colour white is rendered as a positive value, while background values will have a very low value.