Jan 25, 2024

Public workspaceQuantification of area and optical density of intracellular neuromelanin with TruAI in H&E stained sections

DOI
dx.doi.org/10.17504/protocols.io.5qpvo36o7v4o/v1
  • 1Neurodegenerative Diseases Research Group, Vall d’Hebron Research Institute (VHIR)-Network Center for Biomedical Research in Neurodegenerative Diseases (CIBERNED), 08035 Barcelona, Spain;
  • 2Aligning Science Across Parkinson’s (ASAP) Collaborative Research Network, Chevy Chase, MD, 20815;
  • 3Institute of Neurosciences, Autonomous University of Barcelona (INc-UAB), Bellaterra, Barcelona, Spain;
  • 4Catalan Institution for Research and Advanced Studies (ICREA), Barcelona, Spain
Miquel Vila
Open access
DOI: dx.doi.org/10.17504/protocols.io.5qpvo36o7v4o/v1
Protocol Citation: Marta Gonzalez-Sepulveda, Thais Cuadros, Miquel Vila 2024. Quantification of area and optical density of intracellular neuromelanin with TruAI in H&E stained sections. protocols.io https://dx.doi.org/10.17504/protocols.io.5qpvo36o7v4o/v1
License: This is an open access protocol distributed under the terms of the Creative Commons Attribution License,  which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited
Protocol status: Working
We use this protocol and it's working
Created: January 25, 2024
Last Modified: January 25, 2024
Protocol Integer ID: 94116
Funders Acknowledgement:
Aligning Science Across Parkinson’s (ASAP)
Grant ID: ASAP-020505
Abstract
Quantification of area and optical density of intracellular neuromelanin with TruAI in H&E stained sections
Loading training label function in TruAI Software
Loading training label function in TruAI Software
Open scanned images with Olympus VS200 Desktop (EVIDENT Technology GmbH, ver. 4.1.1 build 27564).
Under the ‘Detect’ window, select ‘Training Labels’.
Creating training label classes
Creating training label classes
Create a new training label class by selecting the star icon. Change the class name and the color by double click on it.
A new foreground class will appear under level 1.
Rename as ‘NM’ referring to intracellular neuromelanin.
Repeat steps 3-5 for all the label classes: extracellular NM, Non-pigmented cells and background.
Once you have added all your classes you have to save the set of training classes to use it in several images. It is very important that the classes you use in different images have exactly the same names and colors and the easy way to do it is by saving the set.
Once you have the set you can load it in all your images of interest by clicking the folder icon once the image is opened.
Manage sets of training label classes
Manage sets of training label classes
The hand icon lets you manage the label sets: here you can export a label class to be able to import it in other computer. This is interesting to export a set of label classes from the workstation to your own computer. As we said before, to have exactly the same label classes is very important to analyze a set of images because, if not, the algorithm will not be correctly processed.
Optimize background and all training label classes
Optimize background and all training label classes
Under the automatically created ‘background’ level, select the fill icon and outline an area of the section with approximately 50-100 NM granules. The outlined area must be continuous.
Select the ‘NM’ class and use the same fill option to outline the shape of each NM granule as closely as possible.
To ensure maximum accuracy of the neural training, NM granules of all sizes and densities should be drawn.
Further, all NM granules in each background area should be drawn.
Save and export this NM training class set and apply it to 5-10 scanned sections.
Repeat steps 11-14 identically for each image and for each training class.
Deep learning training
Deep learning training
In the ‘Deep Learning’ window, select ‘New Training’ and pick ‘Image Segmentation’ option.
In the ‘New Training: Input and Output’ pop-up window, load all images used in the ‘Optimize background and all training label classes’ subsection.
Ensure the input channel is RGB.
Select ‘Specific Network (RGB)’ under ‘Training Configuration’.
Start training and run until at least 0.85 similarity is reached.
Applying neural network to the scanned sections:
Applying neural network to the scanned sections:
After successful completion of deep learning training, open a scanned brightfield section with Olympus VS200 Desktop (EVIDENT Technology GmbH, ver. 4.1.1 build 27564).
In the ‘Detect’ window, select the ‘Count and Measure’ drop down menu and pick the ‘New ROI’ option to create ROIs for further anatomical delineation.
Once all the ROIs have been drawn, select the ‘Neural Network Segmentation’ option above.
In the ‘Neural Network Segmentation’ pop-up window, load the saved neural network and adjust the ‘Detection threshold’ to 0%. Proceed by selecting ‘Count and Measure on ROI’.
Thresholding and analysis of the results
Thresholding and analysis of the results
The generated results appear in the ‘Count and Measure Results’.
The corresponding ROI for each NM granule can be found in the ‘ROI’ column. Other computed parameters relevant to size and intensity of NM are also listed, e.g. ‘Area µm2’, ‘Mean (Color Intensity Value)’, ‘Mean (Saturation)’, and ‘Mean (Hue)’.