HTW-KI-Werkstatt/SynthMT

SynthMT是一个合成的、经过专家验证的数据集，旨在对干扰反射显微镜（IRM）样条件下体外微管（MT）图像的分割模型进行基准测试。它提供了真实的合成MT图像、像素级完美的实例分割标签、一个适应任何真实显微镜域而不需要地面真实注释的生成管道，以及对经典、显微镜专用和通用基础模型的全面基准测试。数据集的结构包括唯一的图像标识符、合成的IRM样图像和实例掩码堆栈。微管（MTs）是细胞骨架细丝，对细胞内运输、细胞运动和纺锤体形成至关重要。测量MT的数量、长度和曲率对于体外重组实验、药物发现和机制细胞生物学至关重要。然而，手动MT注释耗时且不可扩展，IRM/TIRF成像在不同实验室之间存在显著差异（域偏移），并且没有大型标记基准用于MT分割。SynthMT直接解决了这一空白。

SynthMT is a synthetic, expert-validated dataset designed to benchmark segmentation models on in vitro microtubule (MT) images visualized in interference reflection microscopy (IRM)–like conditions. It provides realistic synthetic MT images, pixel-perfect instance segmentation labels, a generation pipeline that adapts to any real microscope domain without the need for ground-truth annotations, and a comprehensive benchmark of classical, microscopy-specialized, and general-purpose foundation models. The dataset structure includes a unique image identifier, synthetic IRM-like images, and a stack of instance masks. Microtubules (MTs) are cytoskeletal filaments essential for intracellular transport, cell motility, and mitotic spindle formation. Measuring MT count, length, and curvature is critical for in vitro reconstitution experiments, drug discovery, and mechanistic cell biology. However, manual MT annotation is time-consuming and unscalable, IRM/TIRF imaging varies significantly across labs (domain shift), and no large, labeled benchmarks existed for MT segmentation. SynthMT directly addresses this gap.