Andy’s Algorithms: new automated digital image analysis pipelines for FIJI

Quantification of cellular antigens and their interactions via antibody-based detection methods are widely used in scientific research. Accurate high-throughput quantitation of these assays using general image analysis software can be time consuming and challenging, particularly when attempted by users with limited image processing and analysis knowledge. To overcome this, we have designed Andy’s Algorithms, a series of automated image analysis pipelines for FIJI, that permits rapid, accurate and reproducible batch-processing of 3,3′-diaminobenzidine (DAB) immunohistochemistry, proximity ligation assays (PLAs) and other common assays. Andy’s Algorithms incorporates a step-by-step tutorial and optimization pipeline to make batch image analysis simple for the untrained user and adaptable across laboratories. Andy’s algorithms provide a simpler, faster, standardized work flow compared to existing programs, while offering equivalent performance and additional features, in a free to use open-source application of FIJI. Andy’s Algorithms are available at GitHub, publicly accessed at https://github.com/andlaw1841/Andy-s-Algorithm.

A raw image is selected from the target folder and a color blindness filter (deuteranope filter for total selection or tritanope filter for DAB+ selection) is applied to enhance the selection of the region of interest (ROIs) selection prior to the application of a color deconvolution filter (Feulgen light green filter for total selection or a H&E for DAB+ selection) A Gaussian blur is then applied to the image, which is then converted to an 8-bit grey-scale image before applying a threshold. An optimal automatic threshold function is selected from five main algorithms; (Huang, RenyiEntropy (or Li for the basic pipeline), Otsu, Shanbhag, and Yen). The image is then converted to a binary image and watershed, fill holes, and edge exclusion can be applied before processing with particle exclusion. Image analysis is performed with overlay images of both the total and DAB+ selection produced in the target folder.

Supplementary Figure 2. A new pipeline for H&E image analysis (A)
Flow chart depicting the image processing steps within the H&E particle algorithm for the selection of all hematoxylin (H) rich regions within an H&E image.
A raw image is selected from the target folder and a color blindness filter (deuteranope filter for total tissue selection or tritanope filter for dark blue hematoxylin selection) is applied to enhance region of interest (ROIs) selection prior to the application of a color deconvolution filter (FastRed/Fastblue/DAB filter for total tissue selection or a H&E/DAB for dark blue hematoxylin selection) to discriminate between dark blue and light blue/red. A Gaussian blur is then applied to the image, which is then converted to an 8-bit grey-scale image before applying a threshold. An optimal automatic threshold function is applied (using the calculations Moments, MaxEntropy, Otsu, Triangle (or Intermodes for the hematoxylin dense regions) and Yen for the selection). The image is then converted to a binary image and watershed, fill holes, and edge exclusion can be applied before processing with particle exclusion. Image analysis is performed with overlay images of both the total tissue and dark blue hematoxylin dense selection produced in the target folder. Background is first removed from brightfield images of 3D colony forming assays.
removing shadowing as a result of uneven illumination and shadowing effects (e.g. due to tissue culture wells in 3D colony forming assays, Supp. Fig. 3B). A Gaussian blur is then applied to the image, which is then converted to an 8-bit grey-scale image before applying a threshold. An optimal automatic threshold function is selected from five main algorithms; Huang, Li (or MaxEntropy for the normalize local contrast selection), Otsu, Triangle, and Yen). The image is then converted to a binary image and waterhshed, fill holes, and edge exclusion can be applied before processing with particle exclusion. Image analysis is performed with overlay images of both the total tissue and dark blue hematoxylin dense selection produced in the target folder.   Table 3. Glossary defining the output parameters in the summary spreadsheet for the DAB+ IHC, PLA, H&E and 3D colony forming assay pipelines.

IHC Glossary
Average Intensity The average mean grey value of all positive ROI, that ranges from 0-255 where 0 is darkest (black) and 255 is brightest (white).

Percent Area
Percentage The area of the total selection (total ROI), which can be visualized in the "total selection mask" image

Total Count
The count of the total selection (total ROI), which can be identified in the "total ROI" zip file  The area of the total selection (total ROI), which can be visualized in the "total selection mask" image Total Count The count of the total selection (total ROI), which can be identified in the "total ROI" zip file Total Mask Image Black and white binary image where the total count and area is measured from Total Overlay Image Pseudo-color image with the total ROI overlaid on top of the raw image

3D Colony Assay Glossary
Average Area Per Colony The average area of each colony (total area of all colony divided by colony counts) Average Aspect Ratio Per Colony The average aspect ratio of each colony based on the major axis divided by the minor axis

Average Circularity Per Colony
The average circularity of each colony that ranges from 0-1 where 1 is a perfect circle and 0 is an elongated polygon

Cell Mask
Black and white binary image where the total colony count and area is measured from Cell Overlay Pseudo-color image with the total ROI overlaid on top of the raw image

Colony Counts
The count of total number of colonies, which can be identified in the "cells ROI" zip file ROI Region of Interest

Total Area of All Colony
The area of the total selection, measured based on the "cell masks" image