The kidney is an important organ in the body to excrete metabolic waste, and the glomerulus is an essential structure for the kidney to play a role in blood filtration. Abnormal glomerular numbers are associated with nephropathy or circulatory disease. With the development of imaging technology, mesoscopic optical imaging can obtain whole kidney images at single cell resolution. The detection of glomeruli from images is very crucial for understanding the renal function and studying nephropathy. Existing detection methods cannot balance both accuracy and efficiency, so we proposed a deep learning-based glomeruli detection method. First, we imaged an entire mouse kidney with High-Definition fluorescent Micro-Optical Sectioning Tomography (HD-fMOST) and obtained a three-dimensional (3D) kidney image at cellular resolution. Then, we designed an end-to-end 3D convolutional neural network based on the morphological features of glomeruli in the kidney image, which can directly read 3D images and predict the precise coordinates of glomeruli. We used the acquired kidney dataset to train the network and validated the effect of glomeruli detection. Finally, we applied our approach to detect glomeruli in large-scale mouse kidney. The results showed that the proposed method reached the state-of-the-art level, which is more efficient and accurate compared to similar methods. The proposed method will provide a powerful tool for kidney-related research.
Obtaining fine structures in the whole brain is necessary for understanding brain function. Simple and effective methods for large-scale 3D imaging at optical resolution are still lacking. Here, we proposed a deep-learning-based fluorescence micro-optical sectioning tomography (DL-fMOST) method for fast, high-resolution whole-brain imaging. We utilized a wide-field microscope and a convolutional neural network for optical sectioning imaging, replacing traditional optical method. A 3D dataset of a mouse brain with a voxel size of 0.32 × 0.32 × 2 µm was acquired in 1.5 days. We demonstrated the robustness of DL-fMOST for mouse brains with labeling of different types of neurons.
The brain-wide reconstruction of neuronal population is an indispensible step towards exploring the complete structure of neuronal circuits, a central task that underlies the structure-function relation in neuroscience. Recent advances in molecular labeling and imaging techniques enable us to collect the whole mouse brain imaging dataset at cellular resolution, including the morphological information of neurons across different brain region or even the whole brain. Reconstruction of these neurons poses substantial challenges, and at presents there is no tool for high-speed achieving this reconstruction close to human performance. Here, we presented a tool for filling in the blanks. The tool mainly contains the following function modules: 3D visualization of large-scale imaging dataset, automated reconstruction of neurons, manual editing of the reconstructions at local and global scale. In this tool, in the framework of our previous tools (NeuroGPS-Tree and SparseTracer), the two identifying models were constructed for boosting the automatic level of the reconstruction. One is used to identify the weak signals from inhomogeneous backgrounds and the other is used to identify closely packed neurites. This tool can be suitable for the different big-data formats and can make the dataset be fastly read into memory for the reconstruction. The manual editing module in this tool can correct the errors drawn from above automated algorithms. And thus helps to achieve the reconstruction closer to human performance. We demonstrated the features of our tool on various kinds of sparsely labelled datasets. The results indicated that without loss of the reconstruction accuracy, our tool has a 7-10 folds speed gain over the commercial software that provides the manual reconstruction.
Deciphering the fine morphology and precise location of neurons and neural circuits are crucial to enhance our understanding of brain function and diseases. Traditionally, we have to map brain images to coarse axial-sampling planar reference atlases to orient neural structures. However, this means might fail to orient neural projections at single-cell resolution due to position errors resulting from individual differences at the cellular level. Here, we present a high-throughput imaging method that can automatically obtain the fine morphologies and precise locations of both neurons and circuits, employing wide-field large-volume tomography to acquire three-dimensional images of thick tissue and implementing real-time soma counterstaining to obtain cytoarchitectonic landmarks during the imaging process. The reconstruction and orientation of brain-wide neural circuits at single-neuron resolution can be accomplished for the same mouse brain without additional counterstains or image registration. Using our method, mouse brain imaging datasets of multiple type-specific neurons and circuits were successfully acquired, demonstrating the versatility. The results show that the simultaneous acquisition of labeled neural structures and cytoarchitecture reference at single-neuron resolution in the same brain greatly facilitates precise tracing of long-range projections and accurate locating of nuclei. Our method provides a novel and effective tool for application in studies on genetic dissection, brain function and the pathology of the nervous system.
Neuronal cells play very important role on metabolism regulation and mechanism control, so cell number is a fundamental determinant of brain function. Combined suitable cell-labeling approaches with recently proposed three-dimensional optical imaging techniques, whole mouse brain coronal sections can be acquired with 1-μm voxel resolution. We have developed a completely automatic pipeline to perform cell centroids detection, and provided three-dimensional quantitative information of cells in the primary motor cortex of C57BL/6 mouse. It involves four principal steps: i) preprocessing; ii) image binarization; iii) cell centroids extraction and contour segmentation; iv) laminar density estimation. Investigations on the presented method reveal promising detection accuracy in terms of recall and precision, with average recall rate 92.1% and average precision rate 86.2%. We also analyze laminar density distribution of cells from pial surface to corpus callosum from the output vectorizations of detected cell centroids in mouse primary motor cortex, and find significant cellular density distribution variations in different layers. This automatic cell centroids detection approach will be beneficial for fast cell-counting and accurate density estimation, as time-consuming and error-prone manual identification is avoided.
High-throughput optical imaging is critical to obtain large-scale neural connectivity information of brain in neuroscience. Using a digital mirror device and a scientific complementary metal-oxide semiconductor camera, we report a significant speed improvement of structured illumination microscopy (SIM), which produces a maximum SIM net frame rate of 133 Hz. We perform three-dimensional (3-D) imaging of mouse brain slices at diffraction-limited resolution and demonstrate the fast 3-D imaging capability to a large sample with an imaging rate of 6.9×10 7 pixel/s of our system, an order of magnitude faster than previously reported.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.