2D and 3D Segmentation of uncertain local collagen fiber orientations in SHG microscopy

<strong>General</strong> This dataset consists out of multiple Second Harmonic Generation (SHG) microscopy scans of collagen fibers in mice bones. Some mices are diseased with osteogenesis imperfecta (brittle bone). We used this data to investigate the segmentation of uncertain local collagen fiber orientations. The corresponding paper "2D and 3D Segmentation of uncertain local collagen fiber orientations in SHG microscopy" is accepted at GCPR 2019. <strong>Abstract</strong> Collagen fiber orientations in bones, visible with Second Harmonic Generation (SHG) microscopy, represent the inner structure and its alteration due to influences like cancer. While analyses of these orientations are valuable for medical research, it is not feasible to analyze the needed large amounts of local orientations manually. Since we have uncertain borders for these local orientations only rough regions can be segmented instead of a pixel-wise segmentation. We analyze the effect of these uncertain borders on human performance by a user study. Furthermore, we compare a variety of 2D and 3D methods such as classical approaches like Fourier analysis with state-of-the-art deep neural networks for the classification of local fiber orientations. We present a general way to use pretrained 2D weights in 3D neural networks, such as Inception-ResNet-3D a 3D extension of Inception-ResNet-v2. In a 10 fold cross-validation our two stage segmentation based on Inception-ResNet-3D and transferred 2D ImageNet weights achieves a human comparable accuracy. <strong>Links</strong> A preprint of the paper is available at https://arxiv.org/abs/1907.12868. The final publication is available at Springer via https://doi.org/10.1007/978-3-030-33676-9_26 The source code is available at https://github.com/Emprime/uncertain-fiber-segmentation. <strong>Data description</strong> Please read the accompanying paper for more information about the dataset. Please see the source code for more information about the usage of the data. shg-ce-de: contains the enhanced and denoised scans as image slices, the scans are sorted by mice (wt wildtyp, het ill mice), scan location and individual scan shg-masks: contains the ground truth masks for the three different classes (similar - Green, dissimilar - Red, not of interest - blue) shg-featues: contains the input and gt for the second stage of the proposed two stage segmentation shg-cross-splits: contains the 10 random splits for the 10 fold cross validation logs-prediction: contains the 10 tensorboard logs, weights and predictions for the 10 fold cross validations<strong>概述</strong> 本数据集包含多组针对小鼠骨骼中胶原纤维的二次谐波产生（Second Harmonic Generation, SHG）显微镜扫描图像。部分小鼠患有成骨不全症（脆骨病）。本数据集用于研究不确定局部胶原纤维取向的分割任务，相关论文《2D与3D分割SHG显微镜图像中不确定的局部胶原纤维取向》已被GCPR 2019收录。<strong>摘要</strong> 骨骼中的胶原纤维取向可通过二次谐波产生（SHG）显微镜成像观测，其反映了骨骼内部结构及其在癌症等影响因素下的改变。尽管对这些取向的分析对医学研究具有重要价值，但手动分析所需的大量局部取向并不可行。由于这些局部取向的边界存在不确定性，仅能分割大致区域，而非逐像素分割。我们通过用户研究分析了这些不确定边界对人类标注表现的影响。此外，我们对比了多种二维与三维方法，包括傅里叶分析等经典方法，以及当前最优的用于局部纤维取向分类的深度神经网络。我们提出了一种将预训练二维权重应用于三维神经网络的通用方案，例如Inception-ResNet-3D——这是Inception-ResNet-v2的三维扩展版本。在十折交叉验证中，我们基于Inception-ResNet-3D与迁移的二维ImageNet权重构建的两阶段分割方法，取得了与人类相当的准确率。<strong>链接</strong> 该论文的预印本可在https://arxiv.org/abs/1907.12868获取。最终出版物可通过Springer平台的https://doi.org/10.1007/978-3-030-33676-9_26查阅。源代码可在https://github.com/Emprime/uncertain-fiber-segmentation获取。<strong>数据说明</strong> 有关本数据集的更多信息，请参阅随附的论文。有关数据使用方法的详情，请参考源代码。 shg-ce-de：包含经过增强与去噪的扫描图像切片，扫描图像按小鼠（wt 野生型，het 患病杂合小鼠）、扫描位置与单次扫描进行排序。 shg-masks：包含针对三个不同类别的真值掩码（相似类-绿色、相异类-红色、非关注区域-蓝色）。 shg-featues：包含所提出的两阶段分割方法第二阶段的输入数据与真值标签。 shg-cross-splits：包含用于十折交叉验证的10组随机划分方案。 logs-prediction：包含十折交叉验证对应的10组TensorBoard日志、模型权重与预测结果。