Gut Analysis Toolbox: Training data and 2D models for segmenting enteric neurons, neuronal subtypes and ganglia

This upload is associated with the software, Gut Analysis Toolbox (GAT).

It contains StarDist models for segmenting enteric neurons in 2D, enteric neuronal subtypes in 2D and UNet model for enteric ganglia in 2D in gut wholemount tissue. GAT is implemented in Fiji, but the models can be used in any software that supports StarDist and the use of 2D UNet models. The files here also consist of Python notebooks (Google Colab), training and test data as well as reports on model performance.

The model files are located in the respective folders as zip files. The folders have also been zipped:

• Neuron (Hu; StarDist model):
  • Main folder: 2D_enteric_neuron_model_QA.zip
  • Model File:2D_enteric_neuron_v4_1.zip 
• Neuronal subtype (StarDist model): 
  • Main folder: 2D_enteric_neuron_subtype_model_QA.zip
  • Model File: 2D_enteric_neuron_subtype_v4.zip
• Enteric ganglia (2D UNet model; Use in FIJI with deepImageJ)
  • Main folder: 2D_enteric_ganglia_model_QA.zip
  • Model File:2D_enteric_ganglia_v2.bioimage.io.model.zip

For the all models, files included are:

1. Model for segmenting cells or ganglia in 2D FIJI. StarDist or 2D UNet.
2. Training and Test datasets used for training.
3. Google Colab notebooks used for training and quality assurance (ZeroCost DL4Mic notebooks).
4. Quality assurance reports generated from above notebooks.
5. StarDist model exported for use in QuPath.

The model files can be used within can be used within the software, StarDist. They are intended to be used within FIJI or QuPath, but can be used in any software that supports the implementation of StarDist in 2D.

Data:

All the images were collected from 4 different research labs and a public database (SPARC database) to account for variations in image acquisition, sample preparation and immunolabelling.

For enteric neurons the pan-neuronal marker, Hu has been used and the  2D wholemounts images from mouse, rat and human tissue.

For enteric neuronal subtypes, 2D images for nNOS, MOR, DOR, ChAT, Calretinin, Calbindin, Neurofilament, CGRP and SST from mouse tissue have been used..

25 images were used from the following entries in the SPARC database:

• Howard, M. (2021). 3D imaging of enteric neurons in mouse (Version 1) [Data set]. SPARC Consortium.
• Graham, K. D., Huerta-Lopez, S., Sengupta, R., Shenoy, A., Schneider, S., Wright, C. M., Feldman, M., Furth, E., Lemke, A., Wilkins, B. J., Naji, A., Doolin, E., Howard, M., & Heuckeroth, R. (2020). Robust 3-Dimensional visualization of human colon enteric nervous system without tissue sectioning (Version 1) [Data set]. SPARC Consortium.

The images have been acquired using a combination different microscopes. The images for the mouse tissue were acquired using: 

• Leica TCS-SP8 confocal system (20x HC PL APO NA 1.33, 40 x HC PL APO NA 1.3) 

• Leica TCS-SP8 lightning confocal system (20x HC PL APO NA 0.88) 

• Zeiss Axio Imager M2 (20X HC PL APO NA 0.3) 

• Zeiss Axio Imager Z1 (10X HC PL APO NA 0.45) 

Human tissue images were acquired using: 

• IX71 Olympus microscope (10X HC PL APO NA 0.3) 

For more information, visit: https://github.com/pr4deepr/GutAnalysisToolbox/wiki 

NOTE: The images for enteric neurons and neuronal subtypes have been rescaled to 0.568 µm/pixel for mouse and rat. For human neurons, it has been rescaled to 0.9 µm/pixel . This is to ensure the neuronal cell bodies have similar pixel area across images. The area of cells in pixels can vary based on resolution of image, magnification of objective used, animal species (larger animals -> larger neurons) and potentially how the tissue is stretched during wholemount preparation 

Average neuron area for neuronal model: 701.2 ± 195.9 pixel² (Mean ± SD, 6267 cells)

Average neuron area for neuronal subtype model: 880.9 ± 316 pixel² (Mean ± SD, 924 cells)

Software References:

Stardist

Schmidt, U., Weigert, M., Broaddus, C., & Myers, G. (2018, September). Cell detection with star-convex polygons. In International Conference on Medical Image Computing and Computer-Assisted Intervention (pp. 265-273). Springer, Cham.

deepImageJ

Gómez-de-Mariscal, E., García-López-de-Haro, C., Ouyang, W., Donati, L., Lundberg, E., Unser, M., Muñoz-Barrutia, A. and Sage, D., 2021. DeepImageJ: A user-friendly environment to run deep learning models in ImageJ. Nature Methods, 18(10), pp.1192-1195.

ZeroCost DL4Mic

von Chamier, L., Laine, R.F., Jukkala, J., Spahn, C., Krentzel, D., Nehme, E., Lerche, M., Hernández-Pérez, S., Mattila, P.K., Karinou, E. and Holden, S., 2021. Democratising deep learning for microscopy with ZeroCostDL4Mic. Nature communications, 12(1), pp.1-18.