fluocells (Fluorescent Neuronal Cells)

通过发布这个数据集，我们旨在为使用深度学习的计算机视觉技术提供一个新的测试平台。主要特点是从通用基准数据集的“自然图像”领域向生物成像领域的转变。我们预计这样做的好处可能有两个：i）促进生物医学相关领域的研究——流行的预训练模型通常表现不佳——ii）通过解决这些领域的特殊要求来促进深度学习的方法研究图片。可能的应用包括但不限于语义分割、对象检测和对象计数。数据由通过荧光显微镜获得的小鼠脑切片的 283 张高分辨率图片（1600x1200 像素）组成。最终目标是通过标记对图片中突出显示的神经元进行个体化和计数，从而评估生物实验的结果。相应的真实标签是通过涉及半自动和手动语义分割的混合方法生成的。结果由黑色 (0) 和白色 (255) 图像组成，具有染色神经元所在位置的像素级注释。有关更多信息，请参阅 Morelli, R. 等人，2021。通过使用 c-ResUnet 进行深度学习，在荧光显微镜中自动进行细胞计数。科学报告。 https://doi.org/10.1038/s41598-021-01929-5。原始图像的收集得到了博洛尼亚大学 (RFO 2018) 和欧洲航天局 (研究协议合作 4000123556) 的资助。

By releasing this dataset, we aim to provide a novel testbed for deep learning-based computer vision technologies. Its core characteristic lies in the shift from the domain of "natural images" in general benchmark datasets to the field of biological imaging. We anticipate two key benefits from this shift: i) advancing research in biomedical-related fields, where mainstream pre-trained models typically perform poorly; ii) promoting research on deep learning methodologies by addressing the unique requirements of these domains. Potential applications include, but are not limited to, semantic segmentation, object detection, and object counting. The dataset consists of 283 high-resolution images (1600×1200 pixels) of mouse brain slices acquired via fluorescence microscopy. The ultimate objective is to assess the results of biological experiments by individually identifying and counting the highlighted neurons in the images through annotation. The corresponding ground truth labels were generated using a hybrid approach involving semi-automatic and manual semantic segmentation. The resulting ground truth images consist of black (0) and white (255) binary maps, with pixel-level annotations marking the locations of stained neurons. For more information, please refer to Morelli et al. (2021): Automated cell counting in fluorescence microscopy using deep learning with c-ResUnet. Scientific Reports. https://doi.org/10.1038/s41598-021-01929-5. The collection of the original images was funded by the University of Bologna (RFO 2018) and the European Space Agency (Research Agreement Collaboration 4000123556).

搜集汇总

背景与挑战

背景概述

该数据集包含283张小鼠脑切片的高分辨率荧光显微镜图像，主要用于神经元的语义分割和计数任务，旨在为生物医学图像分析领域的深度学习研究提供测试平台。数据集由博洛尼亚大学于2021年发布，并附有详细的引用论文和来源信息。