Mitoclass

1  Overview

Mitoclass is a napari plugin for the qualitative assessment of mitochondrial network morphology.
Inference is patch‑wise: each 2‑D patch—obtained from a maximum‑intensity projection of 3‑D stacks—is classified as connected, fragmented, or intermediate.

2  Key features

3  Requirements

• Python ≥ 3.10
• OS : Windows, Linux or macOS
• Hardware : CPU is sufficient; GPU (CUDA 11+) is recommended for large datasets

4  Installation

4.1  PyPI

pip install mitoclass

4.2  Reproducible conda environment

conda create -n mitoclass python=3.10
conda activate mitoclass

# (Optional) GPU acceleration
conda install -c conda-forge cudnn=8.9 cuda11.8 tensorflow

pip install mitoclass

Apple Silicon: install tensorflow-macos.

4.3  Pre‑trained model

Download the model (.h5) from
https://github.com/Jmlr2/MitoClassif/releases

5  Usage

5.1  Graphical interface

napari

1. Open Plugins → Mitoclass.

2. Select the four required paths:

   Field	Purpose
   Input dir	Folder of images to analyse (.tif, .tiff, .stk, .png).
   Output dir	Destination folder for CSV and graph files.
   Heatmaps dir	Folder where heatmaps (*_map.tif) will be written.
   Model file	Pre‑trained Keras model (.h5).

3. Click Run inference. A progress bar tracks the number of processed images.

4. After completion:

   • Show heatmaps adds the newly generated *_map.tif layers to napari.
   • Show 3D graph opens graph3d.html, displaying the connected/fragmented/intermediate proportions.

Tip: Without an Input dir you may run Infer active layer; results are still saved to Output dir and Heatmaps dir.

5.2  Output structure

6  Licence

This project is released under the GNU GPL v3 licence.
See the LICENSE file for details.