🔬 LCI Microscopy Course: Image Analysis Workshop

Welcome to the image analysis workshop organized by the Bioimage Informatics facility & the Live Cell Imaging facility as part of the LCI Microscopy course! During this workshop you will:

• See typical examples of what can be done with image analysis and the limitations of each example.

• Understand which image artifacts can be corrected by image analysis in some cases but are easier to correct before acquisition, e.g., uneven illumination, noise.

• Understand which image artifacts cannot be corrected by image analysis, e.g., saturation, underexposure.

To follow up on the examples presented in this workshop, download the code available on this GitHub page (you can do that by clicking on the green <Code> button above). In addition, you also have to download Fiji from the following link. The installation of additional required plugins will be detailed in the following sections.

📂 You can find the datasets for this workshop located in Zenodo. Please download the following:

• Uneven Illumination: img2 and WSI_Brain.zip.

• Handling Noise

• Image Artifacts

🔖 Tip: To open an image in Fiji, go to File > Open. Then, browse the image of interest. If the Bio-Formats plugin interface appears, check the import options and press OK.

🔖 Tip: You can open a script in Fiji by dragging and dropping a file in the main interface or through Plugins > Macros > Edit.... Then, browse the code to be used in each session.

🧑‍🤝‍🧑 Group work: You will be randomly assigned to breakout rooms to work in groups for each activity. Results should be reported in the shared document (link to be sent during the workshop).

⏰ Schedule

• 10:20-11:20 Correction of uneven illumination

• 11:25-12:20 Handling noise

• 12:20-13.20 🥣 Lunch break

• 13:20-14:20 Image artifacts that cannot be fixed

• 14:25-15:20 Nuclei & Cell Segmentation

💪 Let's start!

1) Uneven illumination - how to correct

📂 Dataset: img2 and WSI_Brain.zip.

Uneven illumination can be due to different factors, e.g., paraxial approximation, microscope settings, sample artifacts (samples that are not flat), shading or vignetting (attenuation of the pixel intensity from the centre of the optical axis to the edges).

Uneven illumination can cause discontinuities when stitching whole slide images, background bleaching in time-lapse fluorescent images and compromise downstream analysis when comparing intensities of objects from different regions of the image. In this workshop, we will explore two algorithms that can be used to correct such artifacts.

• Rolling-ball algorithm

• BaSiC plugin (click on the link to follow installation instructions)

➡️ Images to be used in this section are located in within the folder /images/illumination_correction/.

Join your breakout room and then follow the instructions corresponding to your group:

🧑‍🤝‍🧑 Group 1: Illumination correction via rolling ball algorithm

• Open an image from the stitched folder.

• Create a copy of the original image (Image->Duplicate...).

• Go to Process > Subtract Background....

• Choose the radius of the rolling ball algorithm and press OK.

Repeat these steps for the other files in the stitched folder, each corresponding to a different overlap (1%, 5% and 10%). You can also try to correct individual tiles from the tiles folder. Try different values of the radius parameter.

💡 What happens as the size of the radius increases?

💡 What happens as the overlap increases?

🧑‍🤝‍🧑 Groups 2 and 3: Illumination correction via BaSiC plugin

• Install the BaSiC plugin via Fiji update sites (instructions above).

• Open the script Illumination_Basic.ijm found in the scripts folder.

• Update parameter values (lines 12 to 15) according to the instructions below:

  • For img2_ set parameters to:

   overlap = 1; // choose 1, 5 or 10
   grid_size_x = 5;
   grid_size_y = 3;
   file_name = "img";
  

  • For WSI_Brain set parameters to:

   overlap = 10;
   grid_size_x = 9;
   grid_size_y = 6;
   file_name = "C3-BrainSection";
  

• Press the Run button, a Dialog window will appear. Browse the folder named uncorrected inside the reference folder (e.g. ../images/illumination_correction/tiles/img2_1pc/uncorrected/).

  • Group 2: run the script for img2 with different overlap values (img2_1pc, img2_5pc and img2_10pc)

  • Group 3: run the script for img2_10pc and WSI_Brain

Note: Corrected stitched images are saved inside the reference folder

💡 Compare the uncorrected vs corrected images and discuss how the corrected image was improved (or not) for each example.

💡 What happens as the overlap increases? (Group 2)

💡 What happens as the number of tiles increases? (Group 3)

🧑‍🤝‍🧑 Group 4: Uneven focus on z-stacks

• Open the image of interest in Fiji from folder ../images/illumination_correction/uneven_focus/;

• Inspect the z-stack. What is wrong?

• Create a copy of the reference stack (Image > Duplicate, remember to select the Duplicate stack option).

• Segment via thresholding (Image > Adjust > Threshold...). Keep the resulting segmented stack.

• Select the original stack and apply a maximum intensity projection (MIP) via "Image->Stacks->Z project...".

• Create two copies of the projected image.

  • Apply the rolling ball algorithm (Process > Subtract Background...) in one of them.

• Segment both copies via thresholding.

💡 Compare the segmentation results of 1) the z-stack, 2) the MIP, and 3) the corrected MIP.

Note: To help your analysis, you can compare the intensity profiles of the reference images: Draw a straight line from left to right in the reference image, then go to Analyze > Plot Profile.

2) Handling noise

📂 Dataset

Microscopy images may be affected by different types of noise: dark noise from sensors, shot noise (due to the inherent nature of light), and readout noise (due to amplification and conversion of the signal). Different image processing techniques can be used to denoise images. In this workshop, we will focus on convolution filters. The image below shows the effect of the averaging filter during acquisition.

Each group will work with a specific set of images and generate the output measures using the Noise.ijm script.

➡️ Images to be used in this session are located in ../images/noise/. When processing the Tissue_ images, split the channels and select the DAPI channel before running the script (Image > Color > Split Channels).

🧑‍🤝‍🧑 Group assignment:

• All groups: images Nuclei_no_avg.nd2 and Tissue_no_avg.nd2

• Group 1: images Nuclei_avg_2x.nd2 and Tissue_2x_avg.nd2

• Group 2: images Nuclei_avg_4x.nd2 and Tissue_4x_avg.nd2

• Group 3: images Nuclei_avg_8x.nd2 and Tissue_8x_avg.nd2

• Group 4: images Nuclei_avg_16x.nd2 and Tissue_16x_avg.nd2

➡️ Instructions to run the script and generate the measures:

• Open the image of interest in Fiji;

• Open and run the script Noise.ijm;

• Copy the measures from the Summary table to the shared file;

• Inspect the Results table and the Roi Manager tool;

• Take some time to analyze the script!

Testing the Median filter: uncomment line #20 of the script (by removing the // at the beginning of the line; see below). Then, re-run the script for the same reference images. Copy the measures to the shared file and compare the new measures with the previous values.

run("Median...", "radius=4");

💡 What happens as the size of the averaging filter increases?

💡 What happens if we don't use the median filter?

3) Image artifacts

📂 Dataset

Image analysis cannot fix some image artifacts, which may compromise intensity and shape measurements. Examples can be seen in the images below, such as bleedthrough, saturation, under-exposure and soft focus.

3.1) Bleedthrough

3.2) Saturation & Under exposure

3.3) Soft focus

Hands-on exercise

You will work in groups to investigate how each artifact can affect quantification.

➡️ Images to be used in this session are located in ../images/image_artifacts/

🧑‍🤝‍🧑 Group assignment:

• All groups: image ../images/image_artifacts/sequential.nd2

• Group 1: image ../images/image_artifacts/simultaneous.nd2

• Group 2: image ../images/image_artifacts/seq_saturated.nd2

• Group 3: image ../images/image_artifacts/seq_undexposed.nd2

• Group 4: image ../images/image_artifacts/soft_focus.nd2

➡️ Instructions to run the script:

• Open the image of interest in Fiji;

• Open and run the script ImageArtifacts.ijm;

• From the Log window, copy the values of the total segmented area and average intensity of each channel (green and red);

• Take some time to analyze the script!

Wait for all groups to complete the task and add results in the shared file. Then discuss:

💡 How are the measures affected by each artifact?

💡 How do you avoid such artifacts during acquisition?

4) Nuclei & Cell Segmentation

The segmentation of nuclei and cells is a key point of several image analysis tasks in microscopy. In this assignment, we will use Fiji to create a pipeline for segmenting cells and nuclei. Be inspired by the image below!

You can choose to work with your own images acquired during the LCI Microscopy course, or you can work with the images located in the folder ../images/noise/.

➡️ 🔖 Notes & Tips

• Test the different thresholding algorithms in Fiji (Image > Adjust > Threshold...);

• Remember that the background subtraction can improve the segmentation results;

• Pre-processing: you can find several convolution filters in Process > Filters;

• Remember to create a copy of the original image if you want to quantify pixel intensity;

• If thresholding does not perform well, you can try StarDist

  • Available as a plugin for FIJI. Two trained models: fluorescent nuclei and H&E stained images.

• Activate the macro recording function Plugins > Macros > Record... in Fiji to save all the functions and plugins you use. Then, try to create your own script!

Acknowledgements

• 🔬 Image Acquisition by Gabriela Imreh

• 💭 Proposal & Design by Gisele Miranda & Sylvie Le Guyader

• 🛠️ Implementation by Gisele Miranda & Agustin Corbat

• 🎬 Organization by BioImage Informatics facility