Project #14: The speed of life in trees; Linking wood anatomy with wood lifespan and tree growth

In this project, scientist from the Hawkesbury Institute for the Environment, Western Sydney University wanted to investigate how wood anatomy is linked to wood lifespan and tree growth. There are four main cell types in wood: Fiber, Vessel, Axial Parenchyma and Ray. Each type of cell plays a different role in the transport of water and nutrients throughout the plant. By analyzing these cells, scientists hope to gain insights into how they affect wood lifespan and tree growth.
The data for this project were aquired using brightfield microscopy (BFM) of cross-sections of wood samples, which allows researchers to capture images of the entire sample at once. This method provides high-resolution images that can be used to analyze individual cells within the wood structure. The data was included of 93 whole-slide images of 51 species.

Sample data can be downloaded from Zenodo:

Overview

To achieve the desired analysis of the project, as the first step we needed to apply a semantic segmentation model to segment the image into its respective cell types. But the provided data contained ground truth masks only for Vessel cells. Therefore, we provided two different approaches to solve this problem:

• For Vessel cells, we trained a deep learning model to achieve the segmentation masks.
• For other cell types (Fiber, Axial Parenchyma and Ray), we created a pipeline using FeatureForest, a napari plugin that uses a few user-provided scribble labels to generate semantic segmentation masks of desired cell types.

Installation

First you need to install conda or mamba as your python package and dependency manager.
We provided install.sh script for Linux & Mac OS users, and install.bat script for Windows users.
First you need to clone the repo:

git clone https://github.com/ai4life-opencalls/oc_3_projec_14
cd ./oc_3_projec_14

Now run the installation script:

# Linux or Mac OS
sh ./install.sh

# Windows
./install.bat

This will install all required packages and dependencies including napari and the FeatureForest plugin. For more details about installation you can refer to the installation page of the FeatureForest plugin.

Running marimo notebooks:

For various steps we provided a marimo notebook which allows users to have a more interactive interface to run python codes. So, unlike usual jupyter notebooks, the user has no need to edit codes or enters parameters or paths manually.

A marimo notebook is basically a script file that can be executed either by python or marimo command.

marimo is an open-source reactive Python notebook: run a cell or interact with a UI element, and marimo automatically runs dependent cells (or marks them as stale), keeping code and outputs consistent and preventing bugs before they happen. Every marimo notebook is stored as pure Python (Git-friendly), executable as a script, and deployable as an app; while stored as Python, marimo notebooks also have native support for SQL.

To run a notebook interactively, you need to run it using marimo command:

marimo run <notebook_file.py>

This will pop-up a new browser window (tab) where you can interactively execute the cells of the notebook.
If you want to have more control over the notebook execution process, and being able to edit the codes, you need to run it in edit mode:

marimo edit <notebook_file.py>

To exit the interactive session, press Ctrl+C in the console terminal or use the shutdown button in the marimo browser tab.
Here you can find more tutorials on how to work with marimo: https://docs.marimo.io/getting_started/

Semantic Segmentation of Vessel Cells

This pipeline was made for the training of a semantic segmentation model to predict the mask of Vessel cells. You can find the detailed instructions on how to train the model and predict the masks in here

Generating masks for other cell types using FeatureForest

You can find the detailed instructions on how to prepare the data and use FeatureForest in here.

---

AI4Life has received funding from the European Union's Horizon Europe research and innovation programme under grant agreement number 101057970. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Council Executive Agency. Neither the European Union nor the granting authority can be held responsible for them.