Graphical User Interface (GUI)

In this manual, we describe the necessary steps that the User needs to follow to analyse a set of tube images. In addition, the different parameters of the GUI are explained and suggested values are proposed. We tried to make the AutoTube GUI intuitive and easy to use. In the following, the most important steps for the correct usage of the software are explained. Note that each of the steps are associated with a number in red in Figure 2.

Step 1: Click on the “Select Folder” button to locate the directory containing the (*.TIF, *.JPEG, *.PNG) files of your experiments. You should select the top directory containing the images. After that, click “Enter”. You’ll next see that the list of image filenames gets displayed.

Step 2: Microscope Setting parameters. These parameters are related to the microscope and are used to convert distances from pixels to um. There are five main parameters:

• Object Magnification: this is the power of the microscope objective. Example values are: 10x, 20x, and 40x.
• Lens Magnification: extra magnification of e.g. 1.25x, 1.6x, or 2x. Leave it at 1 if your microscope does not have any extra lens magnification.
• C-Mount: is usually set to 1. Other values, e.g. 0.45x, may be used if a camera adapter with that power of magnification is in place.
• Camera Pixel Size (CCD): linear size in um of a physical pixel (assumed to be square) in the CCD chip of the camera used. Typical values are in the range 0.5 - 16 um.
• Binning: combines pixels in the camera to one single pixel in the acquired image - typically used under low light conditions. The common options are 1 (1X1), 2 (2X2), or 4 (4x4). Set this value to 1 as default and it represents that each pixel in the camera is mapped to its own pixel in the output image.

Step 3: Pre-processing parameters for the input images. These parameters are responsible for reducing detrimental effects from image acquisition, such as poor contrast, uneven illumination or noise (For more details, please read the Image Pre-processing subsection in the Materials and Methods Section of the paper). There are four main parameters to be tuned:

• Input Colour Channel: the color channel over which the analysis is going to be performed.
• Adjust Intensity: choose whether an intensity adjustment correction shall be performed. You can choose among three different options: auto-contrast operation, global histogram equalization, and adaptive histogram equalization. The auto-contrast and histogram equalisation options are useful when the contrast across the different image regions is homogeneous, otherwise, use adaptive histogram equalization.
• Correct Illumination: choose whether the illumination should be normalized and set the approximate diameter size (in pixels) associated with the circular uneven illumination effect. Note that this parameter is related to the microscope lenses diameter in pixels. The lenses diameter size that we considered for both lymphatic (LV) and blood vessels (BV) was set to 51, and for the acquisition settings of our images, a value in the range 40-60 pixels.
• Correct Noise: choose whether noise shall be removed from the input images. Options include: BM3D or Wiener Filtering. BM3D is a more robust de-noising method that should be preferred and is the default option for image de-noising. The Wiener filter is a less computationally heavy approach, preferred when results should be obtained faster.

Step 4: Tube-Detection parameters. They are responsible for the actual partition of the images into foreground (vessels) and background regions. For more details, refer to the Tube Detection subsection in the Materials and Methods section of the paper. Five main parameters can be tuned here:

• Detect Finer Tubes: when selected, finer tubes are detected by looking for image regions containing edges. The final image pixel size is given by the following equation: pixel size = camera pixel size * binning / (obj. mag. * lens mag. * c-mount)
• Highlight Tubes: when selected, the Frangi Vesselness Filter is applied. The User also has to give as input an estimate in pixel values of the width of the tubes. This option is especially useful when the quality of the staining is low and also when the images exhibit some artefacts that do not belong to the tubes. We saw empirically that for blood vessels, the values of 3, 5, or 7 obtained good results, while for lymphatic vessels, we used values of 11, 13, or 15.
• Remove Small Regions: after thresholding, regions whose area is below a given percentage of the total image size, are removed. By default, all regions smaller than 1% of the total image size are removed.
• Fill Holes: after thresholding, it is possible to fill holes/gaps in the detected vessels. The User has to define the area in pixels of the holes to be filled.
• Threshold Type: The User can select different Thresholding techniques including, Otsu/Multi- Otsu, Kittler, and Adaptive (CLAHE). The Otsu Thresholding technique is the most conservative Thresholding option. This option is recommended when the images are especially clean and staining is strong. Multi-Otsu is an extension of the Otsu Method, in which additional tubes are detected by considering more pixel classes. In our analysis, we fixed the number of pixel classes/modes to 3, one for the background pixels and the other two for the vessels. Especially when the staining is weak, the third class covers vessel regions that are otherwise not detected. Multi-Otsu is the default option. The Kittler method is sensitive to local noise, however it is able to detect tubes even in weak stained images. Finally, the Adaptive method achieves especially good performance when the images exhibit uneven illumination factors in different regions. For more details and references, please read the Materials & Methods section of the AutoTube manuscript.

Step 5: Tube-Analysis parameters. They are used when computing the morphometric properties of the detected vessels. In specific, they are useful for post-processing the detected tubes.

• Compute Tube Envelope: this option computes the convex-hull over the detected tubes (tube- region envelope) and returns the area of the detected tubes over the area of the envelope. This value is subsequently stored in the output statistical file.
• Remove Short Ramifications: this option removes the ramifications of the skeleton that are shorter than a given pixel value. The User selects the minimum-length for the valid ramifications. For BVs, we found that a good initial value is 30pix and for LV we set this parameter to 15pix, when both images are acquired using a 10x objective.
• Merge Branches Spatially: when the quality of the stainings is low, some image regions appear pixelated. Therefore, the software occasionally detects too many branching points in close proximity. Branching points that are very close together are typically false-positives and therefore can be merged. The User defines the radius of the circular regions over which the branching points are averaged. We found that for BV this parameter can be set up to 15pix (~ 25 um), while for LV we used a parameter value of 10pix (~ 16.7 um). For more details, please read the Tube Analysis subsection in the Materials and Methods Section of the paper.

Step 6: Analysis. Once the parameters are configured and the User clicks on the “Analyse Image” button, the selected images are processed and the morphometric properties are computed. Note that the label of the button gets automatically updated based on the number of images selected, i.e if only one image is selected, then the label of the button is “Analyse 1 Image”, however there is no limit on the total number of images selected for the analysis. Once an image is analysed, then the next image gets automatically processed until all selected images are analysed. On the other hand, a total number of 9 output directories are created on disk to store the output after applying each image processing step. Note that storing the images after each processing step helps the User to have a better understanding of the effect of the parameter values. The User can next adapt the parameters accordingly. Examples of directories include the “01_adjusted” directory, where the images after intensity-adjustment are stored. In the directory “08_overlays”, the final skeleton and branch points are stored for each image. Finally, in the directory “09_statistics”, an excel file is created containing the morphometric properties for each processed image.

Step 7: Manual Correction. It is possible that after the automatic analysis of the images some false- positive or false-negatives tubes might need to be corrected. To do so, the User can paint/add circular regions corresponding to “missing” tubes by left-clicking with the mouse on the “Tube Detection” Image in Tube Detection panel. This will add for each click a circular spot. Also, to remove false-positive tubes, the User needs to right-click with the mouse, the regions that should be removed, on the “Tube Detection” Image in Tube Detection panel. Every time this is executed, the User can adapt the size of the Brush, by changing the parameter value in the Brush-slider.

Step 8: Recompute Skeleton. After having manually corrected the detected tubes, the morphometric tube properties (e.g., skeleton, branch points) need to be recomputed for the new tubes. To do so, the User needs to click on the “Recompute Skeleton” button. Once the processing is finished, the new skeleton and branching-points are displayed in the “Tube Analysis” panel. If needed, the User can modify again the detected tubes. Note that for each single image that needs to be updated, this step needs to be repeated.

Step 9: Statistics. Once the User is satisfied with the detected tubes, the “Statistics” button is clicked to generate a csv file containing the statistics of the morphometric tube properties. The excel file can be found in the “09_statistics” directory that is found in the “output” directory, at the same location, where the images are stored on disk. The name of the file is stats_summary_skelcut{digit1}_branchcut{digit2}.csv The variable digit1 corresponds to the value used for defining the minimum-length of valid ramifications (see Step 4). The variable digit2 corresponds to the value of the radius set to average the branching points (see Step 4).

Note that the csv file will contain the extracted geometrical properties in both pixel and um units. For that purposes, the column names in the csv file will be followed respectively by “(pix)” or “(um)” in their titles to denote the used units.