Synchrotron radiation-based micro computed tomography dataset - in vivo mouse brain - Mouse17

Retrospective cardiac gated imaging at ESRF ID17 beamline This datasets features a synchrotron radiation-based micro computed tomography dataset of a C57BL/6J mouse injected with barium-based contrast agent into the right lateral cerebral ventricle. Data was recorded at the beamline ID17 of the European Synchrotron Radiation Facility in November 2021. For more information, see the associated publication, repository or the FABRIC4 portal. Mouse17_male_live_23p7_0p5ulmin_5ul_60k360_h2_gated010ms.zipFull 3D-stack, 10 ms after the R-peak, 20 - 30 min after start of contrast agent infusion Mouse17_male_live_23p7_0p5ulmin_5ul_60k360_h2_gated010ms_ROI.zipRegion of interest 3D-stack, subvolume of the above dataset containing the nasopharynx Mouse17_male_live_23p7_0p5ulmin_5ul_60k360_h2_gated010ms_ROIsegmentation.zipRegion of interest 3D-stack, segmentation of the nasopharynx in the above dataset Mouse17_male_live_23p7_0p5ulmin_5ul_60k360_h2_gated150ms.zipFull 3D-stack, 150 ms after the R-peak, 20 - 30 min after start of contrast agent infusion Mouse17_male_live_23p7_0p5ulmin_5ul_60k360_h2_gated150ms_ROI.zipRegion of interest 3D-stack, subvolume of the above dataset containing the nasopharynx Mouse17_male_live_23p7_0p5ulmin_5ul_60k360_h2_gated150ms_ROIsegmentation.zipRegion of interest 3D-stack, segmentation of the nasopharynx in the above dataset Mouse17_male_live_23p7_0p5ulmin_5ul_60k360_h2_slice408_gated010-200ms.zip2D slice z = 408, 10-200 ms delay after R-cardiac peak, 20 - 30 min after start of contrast agent infusion, raw data for Supplementary_Video_5.mp4 Supplementary_Video_5.mp42D slice z = 408, 10-200 ms delay after R-cardiac peak, 20 - 30 min after start of contrast agent infusion Due to an imprecise voxel size of 6.0 µm being entered in the metadata and used during reconstruction, the intensity values in the datasets provided here have to be multiplied by 6.0 µm / 6.3 µm = 0.952 for the intensity values to correspond to linear attenuation coefficients. Methods For the dataset shown, a male mouse (SubjectID: Mouse17) twelve weeks of age and 23.7 g body weight was first injected subcutaneously with buprenorphine (0.1 mg/kg) for analgesia. Anesthesia was induced after onset of analgesia 30 min later, via intraperitoneal injection of a cocktail of ketamine (73 mg/kg) and medetomidine (0.18 mg/kg). The depth of anesthesia was monitored by testing reflexes, and additional injections were given as needed. During imaging, this was done via an intraperitoneal infusion line (30G needle with 0.28 mm inner diameter tubing) connected to a remote-controlled syringe pump. Eye ointment was applied and the skull, neck, and upper thoracic region of the mouse were shaved to avoid potential artefacts during X-ray imaging. 2 × 0.5 ml glucose 10% was administered in two separate subcutaneous injections. The mouse was secured in a stereotactic frame without artificial ventilation, and 5 µl of contrast agent were injected into the lateral ventricle using a 34G Hamilton syringe at a flow rate of 0.5 µl/min. Injection coordinates were 0.95 mm lateral and 0.22 mm caudal of the bregma, at 2.3 mm depth. The animal was then transferred to the radiation hutch containing the SRµCT imaging setup described above, where it was imaged with a monochromatic beam at a photon energy of 37.95 keV. 60 000 radiographs over a rotation range of 360° were acquired with a pco.edge 5.5 camera coupled with a Hasselblad 100 mm f/2.2 lens and a 250 µm LuAG:Ce scintillator for 6.3 µm effective pixel size. The field of view was reduced to 2560 × 780 pixels due to the limited height of the X-ray beam, and recorded with 5 ms exposure time and 5 ms overhead time. Acquisition time per scan was 10 min. Sample-detector distance was 3 m. Retrospective cardiac gating was performed using MATLAB (release R2022b; The MathWorks Inc., Natick, USA) based on Fardin et al.. The electrocardiogram (ECG) signal was denoised using a discrete wavelet transform. Detection of ECG peaks was achieved with the findpeaks function, where the parameters minimum peak height and minimum peak distance were optimized via grid search such that the error to the recorded heart rate was minimized. Following peak detection, the time delay between the recorded trigger signal of each acquired projection and the closest R-peak was calculated. The projections were then grouped according to their time delay into time bins of 10 ms width, resulting in 18 bins within the minimum cycle duration and 2 188 projections per bin. For each bin, one tomogram corresponding to a different phase in the cardiac cycle was reconstructed using GPU-accelerated filtered backprojection, implemented in the ASTRA Toolbox (version 2.1.0), and accessed through TomoPy (version 1.12.2). For the initial analysis in 2D, the images of the most decorrelated cardiac phases, i.e. 10 ms and 150 ms after the cardiac R-peak, were subtracted to visually detect geometrical deformations of brain structures. Those structures with an image intensity difference greater than the noise floor could be identified as having moved between the two images. This inspection was performed with both raw and median-filtered images. To quantify the detected motion in 3D, a ROI covering the nasopharynx was cropped from two scans that were cardiac-gated at 10 and 150 ms after the cardiac R-peak. The nasopharynx was segmented using a coarse mask to isolate the nasopharynx from nearby bones, thresholding, morphological operations, and connected component analysis. Specifically, the perimeters of the nasopharynx were extracted by masking the volumes: binarization with a manually selected threshold, morphological closing then opening using a ball with radius 2 pixels, removing regions with volume below 512 voxels corresponding to image noise, closing with a ball of radius 1.5 pixels, extracting the largest component using full connectivity, and finally morphological closing using a ball with radius 2.5 pixels. The resulting perimeter profiles of the nasopharynx were filled using flood filling. These operations were performed in a Python pipeline (version 3.13.1) using scikit-image (version 0.25.0). Nasopharynx surface meshes were created from these segmentations using PyVista (version 0.44.2) and smoothed with a Laplacian filter (1 000 iterations, relaxation factor 0.01) to remove pixelated edges (Fig. 5e). Surface-to-surface distances were calculated for each mesh point in the 150 ms delay dataset by finding the distance to the nearest-neighbor point in the 10 ms delays mesh with a kd-tree (SciPy, version 1.15.1).欧洲同步辐射装置（ESRF）ID17光束线回顾性心脏门控成像数据集 本数据集为基于同步辐射的显微计算机断层扫描（micro-CT）数据集，实验对象为向右侧侧脑室注射钡基造影剂的C57BL/6J小鼠。数据采集于2021年11月，采集地点为欧洲同步辐射装置（ESRF）ID17光束线。如需更多信息，请参阅相关发表文献、代码仓库或FABRIC4门户。 Mouse17_male_live_23p7_0p5ulmin_5ul_60k360_h2_gated010ms.zip：完整3D堆叠数据，采集于心脏R波峰值后10 ms，造影剂注射开始后20~30分钟 Mouse17_male_live_23p7_0p5ulmin_5ul_60k360_h2_gated010ms_ROI.zip：感兴趣区3D堆叠数据，为上述数据集包含鼻咽部的子体积 Mouse17_male_live_23p7_0p5ulmin_5ul_60k360_h2_gated010ms_ROIsegmentation.zip：感兴趣区3D堆叠数据，为上述数据集鼻咽部的分割结果 Mouse17_male_live_23p7_0p5ulmin_5ul_60k360_h2_gated150ms.zip：完整3D堆叠数据，采集于心脏R波峰值后150 ms，造影剂注射开始后20~30分钟 Mouse17_male_live_23p7_0p5ulmin_5ul_60k360_h2_gated150ms_ROI.zip：感兴趣区3D堆叠数据，为上述数据集包含鼻咽部的子体积 Mouse17_male_live_23p7_0p5ulmin_5ul_60k360_h2_gated150ms_ROIsegmentation.zip：感兴趣区3D堆叠数据，为上述数据集鼻咽部的分割结果 Mouse17_male_live_23p7_0p5ulmin_5ul_60k360_h2_slice408_gated010-200ms.zip：z=408位置的2D切片，采集于心脏R波峰值后10~200 ms延迟，造影剂注射开始后20~30分钟，为Supplementary_Video_5.mp4的原始数据 Supplementary_Video_5.mp4：z=408位置的2D切片，采集于心脏R波峰值后10~200 ms延迟，造影剂注射开始后20~30分钟 由于元数据中录入并在重建过程中使用的体素尺寸为不准确的6.0 µm，因此本数据集的强度值需乘以6.0 µm / 6.3 µm = 0.952，才能将其转换为对应的线性衰减系数。 ## 实验方法 针对本数据集的实验对象为12周龄、体重23.7 g的雄性小鼠（受试者编号：Mouse17）：首先经皮下注射丁丙诺啡（0.1 mg/kg）进行镇痛；镇痛起效30分钟后，经腹腔注射氯胺酮（73 mg/kg）与美托咪啶（0.18 mg/kg）的混合制剂诱导麻醉。通过检测反射活动监测麻醉深度，并根据需要追加麻醉剂量。成像期间，通过连接至遥控注射泵的腹腔输注管路（30G针头，内径0.28 mm的管路）维持麻醉状态。为避免X射线成像产生伪影，涂抹眼药膏并剃除小鼠颅骨、颈部及上胸部的毛发。经两次皮下注射给予2×0.5 ml的10%葡萄糖溶液。 将小鼠固定于立体定位架，未进行人工通气；使用34G Hamilton注射器以0.5 µl/min的流速向侧脑室注射5 µl造影剂。注射坐标为：相对于前囟（bregma）向外侧0.95 mm、尾侧0.22 mm，深度2.3 mm。 随后将小鼠转移至装有同步辐射显微CT（SRµCT）成像系统的辐射防护舱，使用37.95 keV的单色X射线束进行成像。采用搭载哈苏100 mm f/2.2镜头与250 µm LuAG:Ce闪烁体的pco.edge 5.5相机采集360°旋转范围内的60000张射线照片，有效像素尺寸为6.3 µm。由于X射线束高度有限，视场被裁剪为2560×780像素，曝光时间为5 ms，额外耗时5 ms。单次扫描的采集时长为10分钟，样品与探测器的距离为3 m。 基于Fardin等人的研究，使用MATLAB（R2022b版本；美国Natick的The MathWorks公司）完成回顾性心脏门控处理。首先通过离散小波变换对心电图（ECG）信号进行去噪；使用findpeaks函数检测ECG峰值，通过网格搜索优化最小峰值高度与最小峰值间距参数，使拟合心率与实测心率的误差最小化。峰值检测完成后，计算每张采集的投影图像的触发信号与最近的R波峰值之间的时间延迟。随后根据时间延迟将投影图像分组至宽度为10 ms的时间窗口中，在最小心动周期内共得到18个窗口，每个窗口包含2188张投影图像。针对每个窗口，使用基于ASTRA Toolbox（2.1.0版本）实现、并通过TomoPy（1.12.2版本）调用的GPU加速滤波反投影算法，重建出对应心动周期不同相位的断层图像。 在初始二维分析中，将心脏R波峰值后10 ms与150 ms两个相关性最弱的心动时相的图像进行相减，以视觉检测脑结构的几何形变。图像强度差异高于噪声基底的结构可被判定为在两幅图像间发生了位移。该目视检查同时使用原始图像与中值滤波后的图像进行。 为量化三维检测到的运动，从心脏R波峰值后10 ms与150 ms门控的两次扫描数据中裁剪出覆盖鼻咽部的感兴趣区。通过粗掩码分离鼻咽部与邻近骨骼、阈值分割、形态学操作及连通分量分析完成鼻咽部的分割。具体流程为：通过手动选取阈值对体积数据进行二值化，使用半径为2像素的球形结构元进行形态学闭运算与开运算，移除体积小于512个体素的图像噪声区域，再使用半径为1.5像素的球形结构元进行闭运算，通过全连通性提取最大连通分量，最后使用半径为2.5像素的球形结构元进行闭运算。通过泛洪填充算法填充提取出的鼻咽部轮廓。上述操作通过基于Python（3.13.1版本）的工作流完成，使用scikit-image（0.25.0版本）实现。使用PyVista（0.44.2版本）从分割结果生成鼻咽部表面网格，并通过拉普拉斯滤波器（1000次迭代，松弛因子0.01）进行平滑以去除像素化边缘（对应图5e）。使用kd树（SciPy，1.15.1版本）计算150 ms延迟数据集中每个网格点与10 ms延迟数据集网格中最近邻点的距离，得到表面间的距离。