Dataset supporting "Nuclear instance segmentation and tracking for preimplantation mouse embryos"

For investigations into fate specification and morphogenesis in time-lapse images of preimplantation embryos, automated 3D instance segmentation of nuclei is invaluable. Supervised machine learning approaches can radically improve segmentation accuracy, but they often require large amounts of annotated 3D data. Using selective plane illumination microscopy (SPIM) we acquired 3D live images of H2B-miRFP720-expressing preimplantation embryos at various developmental stages and created a new ground-truth dataset with full 3D nuclear instance segmentation. This dataset, which we call BlastoSPIM (concatenation of blastocyst and SPIM), is one of the largest and most complete of its kind and can be used for benchmarking different segmentation methods. BlastoSPIM is split into two parts: 1.0, mostly containing images of early embryos, and 2.0, containing images of late blastocysts. The BlastoSPIM 1.0 dataset includes 573 fully annotated 3D images of nuclei in mouse embryos; across all images, there are 11708 individual nuclear instances annotated and 116 annotated polar bodies. The BlastoSPIM 2.0 dataset consists of 80 annotated images of late-stage embryos with a total of 6628 nuclear instances. The quality, detail, and size of the BlastoSPIM dataset makes it unique relative to other publicly available ground truth datasets for nuclear instance segmentation. See blastospim.flatironinstitute.org for more details and models trained on this dataset.
Funding: Research reported in this publication was supported by the National Institutes of Health under award numbers R01HD110577 (E.P.) and R01HD107026 (E.P.) and the National Center for Advancing Translational Sciences (NCATS), a component of the National Institutes of Health (NIH) under award number TL1TR003019 (fellowship, R.P.K.-Y.). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The computations in this work were performed at facilities supported by the Scientific Computing Core at the Flatiron Institute, a division of the Simons Foundation.

为了探究时间推移图像中早期胚胎的命运指定和形态发生，对核的自动三维实例分割显得尤为宝贵。监督式机器学习方法能够显著提升分割精度，但通常需要大量标注的三维数据。通过选择性平面照明显微镜（SPIM）技术，我们获取了表达H2B-miRFP720的早期胚胎在不同发育阶段的3D活体图像，并创建了一个包含完整三维核实例分割的新基准数据集。该数据集被称为BlastoSPIM（融合了囊胚和SPIM），是同类数据集中规模最大、内容最完备的之一，可用于不同分割方法的基准测试。BlastoSPIM分为两个部分：1.0版本，主要包含早期胚胎图像；2.0版本，包含晚期囊胚图像。BlastoSPIM 1.0数据集包括573张完全标注的三维小鼠胚胎核图像；在所有图像中，共有11708个独立的核实例进行了标注，以及116个极体进行了标注。BlastoSPIM 2.0数据集由80张晚期胚胎图像组成，总计包含6628个核实例。BlastoSPIM数据集在质量、细节和规模上相较于其他公开可用的核实例分割基准数据集具有独特优势。更多详细信息及基于该数据集训练的模型，请访问blastospim.flatironinstitute.org。