Abstract:The cytoskeleton is important in controlling the growth and morphology of plant cells, so tracking its morphological changes is essential. Here, we develop a new machine learning based segmentation tool for microtubules (MTs), which can distinguish between polymerised and depolymerised fibres. To circumvent the low abundance of data, we trained on synthetic images of microtubules from a computational micro- tubule model, pre-processed to reproduce microscope effects and partial depolymerisation. We used this tool to investigate how the MT network in an Arabidopsis thaliana root hair cell repolymerises after depoly- merisation under Oryzalin (OZ) drug treatments. Specifically, we show the network initially repolymerises from the shank region. This work demonstrates the viability of using synthetic data to train machine learning systems handling cytoskeletal image data.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **How to effectively track the depolymerization and repolymerization processes of the plant cytoskeleton (especially microtubules)**. Specifically, the author has developed a machine - learning - based segmentation tool that can distinguish between polymerized and depolymerized microtubule fibers and is used to study the changes in the microtubule network in Arabidopsis root - hair cells under drug treatment. ### Problem Background 1. **Importance of the Cytoskeleton** - The cytoskeleton is crucial for the growth and morphological control of plant cells. In particular, microtubules (MTs) are involved in key functions such as cell structure maintenance and cell division. 2. **Existing Challenges** - Existing segmentation tools perform poorly when dealing with depolymerized microtubules, especially in the case of low - data volumes, and it is difficult to accurately detect the depolymerized microtubule residues. - The acquisition of experimental data is costly and time - consuming. Marking data requires manual annotation by experts, which limits the quantity and quality of training data. ### Solutions 1. **Training the Model with Synthetic Data** - To solve the problem of scarce experimental data, the author has developed a pipeline for generating synthetic microtubule images. These images are pre - processed to simulate microscope effects and partial depolymerization phenomena. - The generation of synthetic data includes simulating the growth, contraction, crossing, and bundling behaviors of microtubules, and more realistically reflecting experimental images by adding noise, blurring effects, etc. 2. **Machine - Learning Model** - A convolutional neural network (U - Net architecture) is used for the segmentation of microtubule images. The model can predict whether each pixel belongs to a polymerized microtubule. - The model is trained on a large amount of synthetic data and verified on experimental data, showing good performance, especially when dealing with completely depolymerized and completely polymerized microtubule images. ### Applications and Results 1. **Tracking Microtubule Repolymerization** - The author used this model to study the repolymerization process of the microtubule network in Arabidopsis root - hair cells after Oryzalin drug treatment. - The results show that the microtubule network first starts to repolymerize from the basal region of the root - hair, and the repolymerization area moves as the cell grows. 2. **Verification and Evaluation** - The model performs well on both the control group (untreated) and the completely depolymerized group (after drug treatment) data, especially outperforming traditional methods in terms of performance on depolymerized data. - Due to the lack of reliable benchmark data in the intermediate stages, the author did not verify the data in the partial depolymerization or repolymerization stages. ### Summary This study demonstrates the effectiveness of training a machine - learning model with synthetic data, which can achieve accurate tracking of microtubule depolymerization and repolymerization in the case of scarce experimental data. Future work can be extended to data sets under other microscope conditions and consider applying this method to other cytoskeleton components such as actin.

Machine learning segmentation tool trained on synthetic data for tracking cytoskeleton polymerisation and depolymerisation

Imaging and Reconstruction of Cell Cortex Structures Near the Cell Surface

Deep learning-based cytoskeleton segmentation for accurate high-throughput measurement of cytoskeleton density

Deep machine learning for cell segmentation and quantitative analysis of radial plant growth

Image processing approaches for microtubule remodeling quantification at the immunological synapse

Live Cell Imaging of Microtubule Cytoskeleton and Micromechanical Manipulation of the Arabidopsis Shoot Apical Meristem.

An Automatic Method for Robust and Fast Cell Detection in Bright Field Images from High-Throughput Microscopy.

KnotResolver: Tracking self-intersecting filaments in microscopy using directed graphs

Genetically Encoded Microtubule Binders for Single-Cell Interrogation of Cytoskeleton Dynamics and Protein Activity

A new kymogram-based method reveals unexpected effects of marker protein expression and spatial anisotropy of cytoskeletal dynamics in plant cell cortex

Subcellular structure segmentation from cryo-electron tomograms via machine learning

Multiple Microtubule Tracking in Microscopy Time-Lapse Images Using Piecewise-stationary Multiple Motion Model Kalman Smoother

Machine Learning for Analyzing Atomic Force Microscopy (AFM) Images Generated from Polymer Blends

SpindlesTracker: An Automatic and Low-cost Labeled Workflow for Spindle Analysis

DeepKymoTracker: A tool for accurate construction of cell lineage trees for highly motile cells

Visualizing and Quantifying In Vivo Cortical Cytoskeleton Structure and Dynamics

Spontaneous Downregulation of Antibody/Autoantibody Synthesis in Susceptible Mice upon Chronic Exposure to Mercuric Chloride is not Owing to a General Immunosuppression

Machine learning-guided reconstruction of cytoskeleton network from Live-cell AFM Images

PREVALENCE OF AND MORTALITY FROM HIV TYPE 2 IN GUINEA BISSAU, WEST AFRICA

Automatic microtubule tracking in fluorescence images of cells doped with increasing concentrations of taxol and nocodazole

Visualization of Cortical Microtubule Networks in Plant Cells by Live Imaging and Immunostaining