MTFlow: Time-Conditioned Flow Matching for Microtubule Segmentation in Noisy Microscopy Images

TL;DR

MTFlow addresses the challenge of segmenting thin, curvilinear microtubules in noisy fluorescence images by reframing segmentation as learning a time-dependent vector field $v_\theta(x_t,t)$ that progressively transforms an initial noisy mask toward the ground truth along the path $x_t=(1-t)x_0+tx_1$. The method uses a U-Net backbone with temporal embeddings, computes an Euler-integrated dynamics $x_{n+1}=x_n+\Delta t\,v_\theta(x_n,t_n)$, and outputs the final mask via $\hat{x}=\sigma\big(x_0+\sum_n\Delta t\,v_\theta(\cdot)\big)$. It demonstrates strong segmentation performance on synthetic and real microtubule data, and generalizes to other curvilinear structures such as retinal vessels and corneal nerves (DRIVE, CORN1), outperforming several U-Net-based baselines in accuracy and robustness. The approach offers interpretable, iterative refinement of filament boundaries and shows potential to reduce annotation burden while enabling reliable quantitative analysis of filamentous networks in biomedical imaging.

Abstract

Microtubules are cytoskeletal filaments that play essential roles in many cellular processes and are key therapeutic targets in several diseases. Accurate segmentation of microtubule networks is critical for studying their organization and dynamics but remains challenging due to filament curvature, dense crossings, and image noise. We present MTFlow, a novel time-conditioned flow-matching model for microtubule segmentation. Unlike conventional U-Net variants that predict masks in a single pass, MTFlow learns vector fields that iteratively transport noisy masks toward the ground truth, enabling interpretable, trajectory-based refinement. Our architecture combines a U-Net backbone with temporal embeddings, allowing the model to capture the dynamics of uncertainty resolution along filament boundaries. We trained and evaluated MTFlow on synthetic and real microtubule datasets and assessed its generalization capability on public biomedical datasets of curvilinear structures such as retinal blood vessels and nerves. MTFlow achieves competitive segmentation accuracy comparable to state-of-the-art models, offering a powerful and time-efficient tool for filamentous structure analysis with more precise annotations than manual or semi-automatic approaches.

Figures (5)

• Figure 1: An overview of the proposed MTFlow.
• Figure 2: Segmentation on the MicSim_FluoMT-Simple dataset: (a) Test image, (b) Ground truth, and the composite predictions from (c) MTFlow, and (d) UNet++, in which true positives are in yellow, false positives in red, and false negatives in green.
• Figure 3: Segmentation on the MicSim_FluoMT-Complex dataset: (a) Test image, (b) Ground truth, and the composite predictions from (c) MTFlow, and (d) U-Net++, in which true positives are in yellow, false positives in red, and false negatives in green.
• Figure 4: Segmentation on the MicReal_FluoMT dataset: (a) Test image, (b) Ground truth, and the predictions from (c) MTFlow, and (d) U-Net++.
• Figure 5: Segmentation on the DRIVE (a-d) and the CORN1(e-h) datasets: (a,e) Test image, (b,f) Ground truth, and the predictions from (c,g) MTFlow, and (d,h) U-Net++.