MicSim_FluoMT: Two synthetic datasets of images of fluorescent microtubules (Ait Laydi et al., 2026)

These datasets support the paper by Ait Laydi et al. (2026). They consist of two synthetic image datasets that mimic microscopy images of fluorescently labeled microtubules. These datasets were used to train various deep learning architectures for segmenting microtubules. They include both the input images (referred to as 'noisy') and their corresponding ground truth images (referred to as 'binary'). The 'easy' dataset contains 1192 images where the fluorescence along each microtubule is uniform. The 'hard' dataset also contains 1192 images with the same ground truth as the 'easy' dataset, but the fluorescence along each microtubule decreases towards the extremities. As a result, segmenting the microtubules is more challenging in the 'hard' dataset, particularly at the extremities."

The images were generated using Cytosim [1] and confocalGN [2] softwares:

-       https://github.com/SergeDmi/ConfocalGN

-       https://gitlab.com/f-nedelec/cytosim

Please cite the reference below when using the dataset:

Ait Laydi et al., "A novel attention mechanism for noise-adaptive and robust segmentation of microtubules in microscopy images", 2026. BioRxiv. DOI: 10.1101/2025.10.23.684152

This work was carried out as part of a collaborative project between Hélène Bouvrais from IGDR (CNRS, University of Rennes, France) and Yousef El Mourabit from TIAD Laboratory, FST_BM (Sultan Moulay Slimane University, Morocco).