Analyzing Image Segmentation for Connectomics

Local shape descriptors for neuron segmentation

We present an auxiliary learning task for the problem of neuron segmentation in electron microscopy volumes. The auxiliary task consists of the prediction of local shape descriptors (LSDs), which we combine with conventional voxel-wise direct neighbor affinities for neuron boundary detection. The shape descriptors capture local statistics about the neuron to be segmented, such as diameter, elongation, and direction. On a study comparing several existing methods across various specimen, imaging techniques, and resolutions, auxiliary learning of LSDs consistently increases segmentation accuracy of affinity-based methods over a range of metrics. Furthermore, the addition of LSDs promotes affinity-based segmentation methods to be on par with the current state of the art for neuron segmentation (flood-filling networks), while being two orders of magnitudes more efficient-a critical requirement for the processing of future petabyte-sized datasets.

Scalable Biologically-Aware Skeleton Generation for Connectomic Volumes

As connectomic datasets exceed hundreds of terabytes in size, accurate and efficient skeleton generation of the label volumes has evolved into a critical component of the computation pipeline used for analysis, evaluation, visualization, and error correction. We propose a novel topological thinning strategy that uses biological-constraints to produce accurate centerlines from segmented neuronal volumes while still maintaining biologically relevant properties. Current methods are either agnostic to the underlying biology, have non-linear running times as a function of the number of input voxels, or both. First, we eliminate from the input segmentation biologically-infeasible bubbles, pockets of voxels incorrectly labeled within a neuron, to improve segmentation accuracy, allow for more accurate centerlines, and increase processing speed. Next, a Convolutional Neural Network (CNN) detects cell bodies from the input segmentation, allowing us to anchor our skeletons to the somata. Lastly, a synapse-aware topological thinning approach produces expressive skeletons for each neuron with a nearly one-to-one correspondence between endpoints and synapses. We simultaneously estimate geometric properties of neurite width and geodesic distance between synapse and cell body, improving accuracy by 47.5% and 62.8% over baseline methods. We separate the skeletonization process into a series of computation steps, leveraging data-parallel strategies to increase throughput significantly. We demonstrate our results on over 1250 neurons and neuron fragments from three different species, processing over one million voxels per second per CPU with linear scalability.

Computed Tomography Image Segmentation Using Edge Correction Algorithm for Refractory Mycoplasma Pneumonia in Children

Objective. It is to study the application of edge correction algorithm (ECA) in computed tomography (CT) medical image segmentation, explore its guiding significance in the analysis of clinical characteristics of children with refractory mycoplasma pneumoniae (RMPP), and discuss the therapeutic value of fiberoptic bronchoscopy bronchoalveolar lavage (BAL) for RMPP. Methods. The accuracy of ECA in CT medical image segmentation of children with RMPP was compared with that of the watershed segmentation algorithm (WSA) and swarm intelligence optimization algorithm (SIOA). The clinical characteristics and the imaging characteristics of 80 children with RMPP admitted to hospital from January 2018 to January 2020 were retrospectively analyzed based on the ECA. All children were divided into a lavage group (BAL group, n = 69) and a nonlavage group (non-BAL group, n = 11) according to whether fiberoptic bronchoscopy and BAL were performed. Bronchoscopy was adopted to analyze the cytological characteristics of BAL fluid (BALF) in children, and the recovery rate and the total effective rate of the two groups of children were observed and compared. Results. The overall accuracies (OAs) of the three ECAs (Roberts operator (RO), Sobel operator (SO), and Prewitt operator (PO)) were higher than that of WSA and SIOA, their false negative rate (FNR) and false positive rate (FPR) were small, and their denoising performance was superior to that of WSA and SIOA. The main clinical manifestations of all children were high fever, irritating dry cough, and few early signs. The results of chest CT examination were mainly manifested as patchy or large-scale consolidation, two lung mesh or small nodular shadows, and atelectasis. 69 cases with fiberoptic bronchoscopy showed swelling and congestion of the bronchial mucosa at the lesion site with visible viscous secretions, which was consistent with the imaging changes. The total number of cells in the BALF of children increased ( $P<0.05$ ), which mainly represented the increase of neutrophils ( $P<0.05$ ). The recovery rate of children with lavage (81.16%) was higher dramatically than that of the nonlavage group (45.45%). Conclusion. The ECA had good accuracy and denoising performance in lung CT image segmentation. The clinical characteristics, imaging characteristics, and cytological components of children had changed when they suffered from the RMPP, and fiberoptic bronchoscopy lavage had a therapeutic effect on it.

Deep learning for automatic segmentation of the nuclear envelope in electron microscopy data, trained with volunteer segmentations

Advancements in volume electron microscopy mean it is now possible to generate thousands of serial images at nanometre resolution overnight, yet the gold standard approach for data analysis remains manual segmentation by an expert microscopist, resulting in a critical research bottleneck. Although some machine learning approaches exist in this domain, we remain far from realizing the aspiration of a highly accurate, yet generic, automated analysis approach, with a major obstacle being lack of sufficient high-quality ground-truth data. To address this, we developed a novel citizen science project, Etch a Cell, to enable volunteers to manually segment the nuclear envelope (NE) of HeLa cells imaged with serial blockface scanning electron microscopy. We present our approach for aggregating multiple volunteer annotations to generate a high-quality consensus segmentation and demonstrate that data produced exclusively by volunteers can be used to train a highly accurate machine learning algorithm for automatic segmentation of the NE, which we share here, in addition to our archived benchmark data.

CEM500K, a large-scale heterogeneous unlabeled cellular electron microscopy image dataset for deep learning

Automated segmentation of cellular electron microscopy (EM) datasets remains a challenge. Supervised deep learning (DL) methods that rely on region-of-interest (ROI) annotations yield models that fail to generalize to unrelated datasets. Newer unsupervised DL algorithms require relevant pre-training images, however, pre-training on currently available EM datasets is computationally expensive and shows little value for unseen biological contexts, as these datasets are large and homogeneous. To address this issue, we present CEM500K, a nimble 25 GB dataset of 0.5 × 106 unique 2D cellular EM images curated from nearly 600 three-dimensional (3D) and 10,000 two-dimensional (2D) images from >100 unrelated imaging projects. We show that models pre-trained on CEM500K learn features that are biologically relevant and resilient to meaningful image augmentations. Critically, we evaluate transfer learning from these pre-trained models on six publicly available and one newly derived benchmark segmentation task and report state-of-the-art results on each. We release the CEM500K dataset, pre-trained models and curation pipeline for model building and further expansion by the EM community. Data and code are available at https://www.ebi.ac.uk/pdbe/emdb/empiar/entry/10592/ and https://git.io/JLLTz.

Anatomy and activity patterns in a multifunctional motor neuron and its surrounding circuits

Dorsal Excitor motor neuron DE-3 in the medicinal leech plays three very different dynamical roles in three different behaviors. Without rewiring its anatomical connectivity, how can a motor neuron dynamically switch roles to play appropriate roles in various behaviors? We previously used voltage-sensitive dye imaging to record from DE-3 and most other neurons in the leech segmental ganglion during (fictive) swimming, crawling, and local-bend escape (Tomina and Wagenaar, 2017). Here, we repeated that experiment, then re-imaged the same ganglion using serial blockface electron microscopy and traced DE-3's processes. Further, we traced back the processes of DE-3's presynaptic partners to their respective somata. This allowed us to analyze the relationship between circuit anatomy and the activity patterns it sustains. We found that input synapses important for all the behaviors were widely distributed over DE-3's branches, yet that functional clusters were different during (fictive) swimming vs. crawling.

Transforming the development and dissemination of cutting-edge microscopy and computation

We propose a network of National Imaging Centers that provide collaborative, interdisciplinary spaces needed for developing, applying and teaching advanced biological imaging techniques. Our proposal is based on recommendations from an NSF sponsored workshop on realizing the promise of innovations in imaging and computation for biological discovery.