Large Scale Image Segmentation with Structured Loss Based Deep Learning for Connectome Reconstruction

A robust transformer-based pipeline of 3D cell alignment, denoise and instance segmentation on electron microscopy sequence images

Germline cells are critical for transmitting genetic information to subsequent generations in biological organisms. While their differentiation from somatic cells during embryonic development is well-documented in most animals, the regulatory mechanisms initiating plant germline cells are not well understood. To thoroughly investigate the complex morphological transformations of their ultrastructure over developmental time, nanoscale 3D reconstruction of entire plant tissues is necessary, achievable exclusively through electron microscopy imaging. This paper presents a full-process framework designed for reconstructing large-volume plant tissue from serial electron microscopy images. The framework ensures end-to-end direct output of reconstruction results, including topological networks and morphological analysis. The proposed 3D cell alignment, denoise, and instance segmentation pipeline (3DCADS) leverages deep learning to provide a cell instance segmentation workflow for electron microscopy image series, ensuring accurate and robust 3D cell reconstructions with high computational efficiency. The pipeline involves five stages: the registration of electron microscopy serial images; image enhancement and denoising; semantic segmentation using a Transformer-based neural network; instance segmentation through a supervoxel-based clustering algorithm; and an automated analysis and statistical assessment of the reconstruction results, with the mapping of topological connections. The 3DCADS model's precision was validated on a plant tissue ground-truth dataset, outperforming traditional baseline models and deep learning baselines in overall accuracy. The framework was applied to the reconstruction of early meiosis stages in the anthers of Arabidopsis thaliana, resulting in a topological connectivity network and analysis of morphological parameters and characteristics of cell distribution. The experiment underscores the 3DCADS model's potential for biological tissue identification and its significance in quantitative analysis of plant cell development, crucial for examining samples across different genetic phenotypes and mutations in plant development. Additionally, the paper discusses the regulatory mechanisms of Arabidopsis thaliana's germline cells and the development of stamen cells before meiosis, offering new insights into the transition from somatic to germline cell fate in plants.

RoboEM: automated 3D flight tracing for synaptic-resolution connectomics

Mapping neuronal networks from three-dimensional electron microscopy (3D-EM) data still poses substantial reconstruction challenges, in particular for thin axons. Currently available automated image segmentation methods require manual proofreading for many types of connectomic analysis. Here we introduce RoboEM, an artificial intelligence-based self-steering 3D 'flight' system trained to navigate along neurites using only 3D-EM data as input. Applied to 3D-EM data from mouse and human cortex, RoboEM substantially improves automated state-of-the-art segmentations and can replace manual proofreading for more complex connectomic analysis problems, yielding computational annotation cost for cortical connectomes about 400-fold lower than the cost of manual error correction.

Automatic vectorization of historical maps: A benchmark

Shape vectorization is a key stage of the digitization of large-scale historical maps, especially city maps that exhibit complex and valuable details. Having access to digitized buildings, building blocks, street networks and other geographic content opens numerous new approaches for historical studies such as change tracking, morphological analysis and density estimations. In the context of the digitization of Paris atlases created in the 19th and early 20th centuries, we have designed a supervised pipeline that reliably extract closed shapes from historical maps. This pipeline is based on a supervised edge filtering stage using deep filters, and a closed shape extraction stage using a watershed transform. It relies on probable multiple suboptimal methodological choices that hamper the vectorization performances in terms of accuracy and completeness. Objectively investigating which solutions are the most adequate among the numerous possibilities is comprehensively addressed in this paper. The following contributions are subsequently introduced: (i) we propose an improved training protocol for map digitization; (ii) we introduce a joint optimization of the edge detection and shape extraction stages; (iii) we compare the performance of state-of-the-art deep edge filters with topology-preserving loss functions, including vision transformers; (iv) we evaluate the end-to-end deep learnable watershed against Meyer watershed. We subsequently design the critical path for a fully automatic extraction of key elements of historical maps. All the data, code, benchmark results are freely available at https://github.com/soduco/Benchmark_historical_map_vectorization.

Isotropic multi-scale neuronal reconstruction from high-ratio expansion microscopy with contrastive unsupervised deep generative models

BACKGROUND AND OBJECTIVE

Current methods for imaging reconstruction from high-ratio expansion microscopy (ExM) data are limited by anisotropic optical resolution and the requirement for extensive manual annotation, creating a significant bottleneck in the analysis of complex neuronal structures.

METHODS

We devised an innovative approach called the IsoGAN model, which utilizes a contrastive unsupervised generative adversarial network to sidestep these constraints. This model leverages multi-scale and isotropic neuron/protein/blood vessel morphology data to generate high-fidelity 3D representations of these structures, eliminating the need for rigorous manual annotation and supervision. The IsoGAN model introduces simplified structures with idealized morphologies as shape priors to ensure high consistency in the generated neuronal profiles across all points in space and scalability for arbitrarily large volumes.

RESULTS

The efficacy of the IsoGAN model in accurately reconstructing complex neuronal structures was quantitatively assessed by examining the consistency between the axial and lateral views and identifying a reduction in erroneous imaging artifacts. The IsoGAN model accurately reconstructed complex neuronal structures, as evidenced by the consistency between the axial and lateral views and a reduction in erroneous imaging artifacts, and can be further applied to various biological samples.

CONCLUSION

With its ability to generate detailed 3D neurons/proteins/blood vessel structures using significantly fewer axial view images, IsoGAN can streamline the process of imaging reconstruction while maintaining the necessary detail, offering a transformative solution to the existing limitations in high-throughput morphology analysis across different structures.

Expanded tube attention for tubular structure segmentation

PURPOSE

Semantic segmentation of tubular structures, such as blood vessels and cell membranes, is a very difficult task, and it tends to break many predicted regions in the middle. This problem is due to the fact that tubular ground truth is very thin, and the number of pixels is extremely unbalanced compared to the background.

METHODS

We present a novel training method using pseudo-labels generated by morphological transformation. Furthermore, we present an attention module using thickened pseudo-labels, called the expanded tube attention (ETA) module. By using the ETA module, the network learns thickened regions based on pseudo-labels at first and then gradually learns thinned original regions while transferring information in the thickened regions as an attention map.

RESULTS

Through experiments conducted on retina vessel image datasets using various evaluation measures, we confirmed that the proposed method using ETA modules improved the clDice metric accuracy in comparison with the conventional methods.

CONCLUSIONS

We demonstrated that the proposed novel expanded tube attention module using thickened pseudo-labels can achieve easy-to-hard learning.

Current Progress and Challenges in Large-Scale 3D Mitochondria Instance Segmentation

In this paper, we present the results of the MitoEM challenge on mitochondria 3D instance segmentation from electron microscopy images, organized in conjunction with the IEEE-ISBI 2021 conference. Our benchmark dataset consists of two large-scale 3D volumes, one from human and one from rat cortex tissue, which are 1,986 times larger than previously used datasets. At the time of paper submission, 257 participants had registered for the challenge, 14 teams had submitted their results, and six teams participated in the challenge workshop. Here, we present eight top-performing approaches from the challenge participants, along with our own baseline strategies. Posterior to the challenge, annotation errors in the ground truth were corrected without altering the final ranking. Additionally, we present a retrospective evaluation of the scoring system which revealed that: 1) challenge metric was permissive with the false positive predictions; and 2) size-based grouping of instances did not correctly categorize mitochondria of interest. Thus, we propose a new scoring system that better reflects the correctness of the segmentation results. Although several of the top methods are compared favorably to our own baselines, substantial errors remain unsolved for mitochondria with challenging morphologies. Thus, the challenge remains open for submission and automatic evaluation, with all volumes available for download.

Multi-layered maps of neuropil with segmentation-guided contrastive learning

Maps of the nervous system that identify individual cells along with their type, subcellular components and connectivity have the potential to elucidate fundamental organizational principles of neural circuits. Nanometer-resolution imaging of brain tissue provides the necessary raw data, but inferring cellular and subcellular annotation layers is challenging. We present segmentation-guided contrastive learning of representations (SegCLR), a self-supervised machine learning technique that produces representations of cells directly from 3D imagery and segmentations. When applied to volumes of human and mouse cortex, SegCLR enables accurate classification of cellular subcompartments and achieves performance equivalent to a supervised approach while requiring 400-fold fewer labeled examples. SegCLR also enables inference of cell types from fragments as small as 10 μm, which enhances the utility of volumes in which many neurites are truncated at boundaries. Finally, SegCLR enables exploration of layer 5 pyramidal cell subtypes and automated large-scale analysis of synaptic partners in mouse visual cortex.

Three-dimensional reconstructions of mechanosensory end organs suggest a unifying mechanism underlying dynamic, light touch

Across mammalian skin, structurally complex and diverse mechanosensory end organs respond to mechanical stimuli and enable our perception of dynamic, light touch. How forces act on morphologically dissimilar mechanosensory end organs of the skin to gate the requisite mechanotransduction channel Piezo2 and excite mechanosensory neurons is not understood. Here, we report high-resolution reconstructions of the hair follicle lanceolate complex, Meissner corpuscle, and Pacinian corpuscle and the subcellular distribution of Piezo2 within them. Across all three end organs, Piezo2 is restricted to the sensory axon membrane, including axon protrusions that extend from the axon body. These protrusions, which are numerous and elaborate extensively within the end organs, tether the axon to resident non-neuronal cells via adherens junctions. These findings support a unified model for dynamic touch in which mechanical stimuli stretch hundreds to thousands of axon protrusions across an end organ, opening proximal, axonal Piezo2 channels and exciting the neuron.

Segmentation in large-scale cellular electron microscopy with deep learning: A literature survey

Electron microscopy (EM) enables high-resolution imaging of tissues and cells based on 2D and 3D imaging techniques. Due to the laborious and time-consuming nature of manual segmentation of large-scale EM datasets, automated segmentation approaches are crucial. This review focuses on the progress of deep learning-based segmentation techniques in large-scale cellular EM throughout the last six years, during which significant progress has been made in both semantic and instance segmentation. A detailed account is given for the key datasets that contributed to the proliferation of deep learning in 2D and 3D EM segmentation. The review covers supervised, unsupervised, and self-supervised learning methods and examines how these algorithms were adapted to the task of segmenting cellular and sub-cellular structures in EM images. The special challenges posed by such images, like heterogeneity and spatial complexity, and the network architectures that overcame some of them are described. Moreover, an overview of the evaluation measures used to benchmark EM datasets in various segmentation tasks is provided. Finally, an outlook of current trends and future prospects of EM segmentation is given, especially with large-scale models and unlabeled images to learn generic features across EM datasets.

uRP: An integrated research platform for one-stop analysis of medical images

Introduction

Medical image analysis is of tremendous importance in serving clinical diagnosis, treatment planning, as well as prognosis assessment. However, the image analysis process usually involves multiple modality-specific software and relies on rigorous manual operations, which is time-consuming and potentially low reproducible.

Methods

We present an integrated platform - uAI Research Portal (uRP), to achieve one-stop analyses of multimodal images such as CT, MRI, and PET for clinical research applications. The proposed uRP adopts a modularized architecture to be multifunctional, extensible, and customizable.

Results and Discussion

The uRP shows 3 advantages, as it 1) spans a wealth of algorithms for image processing including semi-automatic delineation, automatic segmentation, registration, classification, quantitative analysis, and image visualization, to realize a one-stop analytic pipeline, 2) integrates a variety of functional modules, which can be directly applied, combined, or customized for specific application domains, such as brain, pneumonia, and knee joint analyses, 3) enables full-stack analysis of one disease, including diagnosis, treatment planning, and prognosis assessment, as well as full-spectrum coverage for multiple disease applications. With the continuous development and inclusion of advanced algorithms, we expect this platform to largely simplify the clinical scientific research process and promote more and better discoveries.

Three-dimensional reconstructions of mechanosensory end organs suggest a unifying mechanism underlying dynamic, light touch

Specialized mechanosensory end organs within mammalian skin-hair follicle-associated lanceolate complexes, Meissner corpuscles, and Pacinian corpuscles-enable our perception of light, dynamic touch 1 . In each of these end organs, fast-conducting mechanically sensitive neurons, called Aβ low-threshold mechanoreceptors (Aβ LTMRs), associate with resident glial cells, known as terminal Schwann cells (TSCs) or lamellar cells, to form complex axon ending structures. Lanceolate-forming and corpuscle-innervating Aβ LTMRs share a low threshold for mechanical activation, a rapidly adapting (RA) response to force indentation, and high sensitivity to dynamic stimuli 1-6 . How mechanical stimuli lead to activation of the requisite mechanotransduction channel Piezo2 7-15 and Aβ RA-LTMR excitation across the morphologically dissimilar mechanosensory end organ structures is not understood. Here, we report the precise subcellular distribution of Piezo2 and high-resolution, isotropic 3D reconstructions of all three end organs formed by Aβ RA-LTMRs determined by large volume enhanced Focused Ion Beam Scanning Electron Microscopy (FIB-SEM) imaging. We found that within each end organ, Piezo2 is enriched along the sensory axon membrane and is minimally or not expressed in TSCs and lamellar cells. We also observed a large number of small cytoplasmic protrusions enriched along the Aβ RA-LTMR axon terminals associated with hair follicles, Meissner corpuscles, and Pacinian corpuscles. These axon protrusions reside within close proximity to axonal Piezo2, occasionally contain the channel, and often form adherens junctions with nearby non-neuronal cells. Our findings support a unified model for Aβ RA-LTMR activation in which axon protrusions anchor Aβ RA-LTMR axon terminals to specialized end organ cells, enabling mechanical stimuli to stretch the axon in hundreds to thousands of sites across an individual end organ and leading to activation of proximal Piezo2 channels and excitation of the neuron.

Deep neural network automated segmentation of cellular structures in volume electron microscopy

Volume electron microscopy is an important imaging modality in contemporary cell biology. Identification of intracellular structures is a laborious process limiting the effective use of this potentially powerful tool. We resolved this bottleneck with automated segmentation of intracellular substructures in electron microscopy (ASEM), a new pipeline to train a convolutional neural network to detect structures of a wide range in size and complexity. We obtained dedicated models for each structure based on a small number of sparsely annotated ground truth images from only one or two cells. Model generalization was improved with a rapid, computationally effective strategy to refine a trained model by including a few additional annotations. We identified mitochondria, Golgi apparatus, endoplasmic reticulum, nuclear pore complexes, caveolae, clathrin-coated pits, and vesicles imaged by focused ion beam scanning electron microscopy. We uncovered a wide range of membrane-nuclear pore diameters within a single cell and derived morphological metrics from clathrin-coated pits and vesicles, consistent with the classical constant-growth assembly model.

Pumping up the volume

The time and cost of annotating ground-truth images and network training are major challenges to utilizing machine learning to automate the mining of volume electron microscopy data. In this issue, Gallusser et al. (2023. J. Cell Biol.https://doi.org/10.1083/jcb.202208005) present a less computationally intense pipeline to detect a single type of organelle using a limited number of loosely annotated images.

Instance segmentation of mitochondria in electron microscopy images with a generalist deep learning model trained on a diverse dataset

Mitochondria are extremely pleomorphic organelles. Automatically annotating each one accurately and precisely in any 2D or volume electron microscopy (EM) image is an unsolved computational challenge. Current deep learning-based approaches train models on images that provide limited cellular contexts, precluding generality. To address this, we amassed a highly heterogeneous ∼1.5 × 10⁶ image 2D unlabeled cellular EM dataset and segmented ∼135,000 mitochondrial instances therein. MitoNet, a model trained on these resources, performs well on challenging benchmarks and on previously unseen volume EM datasets containing tens of thousands of mitochondria. We release a Python package and napari plugin, empanada, to rapidly run inference, visualize, and proofread instance segmentations. A record of this paper's transparent peer review process is included in the supplemental information.

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.

Structured cerebellar connectivity supports resilient pattern separation

The cerebellum is thought to help detect and correct errors between intended and executed commands^1,2 and is critical for social behaviours, cognition and emotion^3-6. Computations for motor control must be performed quickly to correct errors in real time and should be sensitive to small differences between patterns for fine error correction while being resilient to noise⁷. Influential theories of cerebellar information processing have largely assumed random network connectivity, which increases the encoding capacity of the network's first layer^8-13. However, maximizing encoding capacity reduces the resilience to noise⁷. To understand how neuronal circuits address this fundamental trade-off, we mapped the feedforward connectivity in the mouse cerebellar cortex using automated large-scale transmission electron microscopy and convolutional neural network-based image segmentation. We found that both the input and output layers of the circuit exhibit redundant and selective connectivity motifs, which contrast with prevailing models. Numerical simulations suggest that these redundant, non-random connectivity motifs increase the resilience to noise at a negligible cost to the overall encoding capacity. This work reveals how neuronal network structure can support a trade-off between encoding capacity and redundancy, unveiling principles of biological network architecture with implications for the design of artificial neural networks.

Volume electron microscopy reveals age-related circuit remodeling in the auditory brainstem

The medial nucleus of the trapezoid body (MNTB) is an integral component of the auditory brainstem circuitry involved in sound localization. The giant presynaptic nerve terminal with multiple active zones, the calyx of Held (CH), is a hallmark of this nucleus, which mediates fast and synchronized glutamatergic synaptic transmission. To delineate how these synaptic structures adapt to reduced auditory afferents due to aging, we acquired and reconstructed circuitry-level volumes of mouse MNTB at different ages (3 weeks, 6, 18, and 24 months) using serial block-face electron microscopy. We used C57BL/6J, the most widely inbred mouse strain used for transgenic lines, which displays a type of age-related hearing loss. We found that MNTB neurons reduce in density with age. Surprisingly we observed an average of approximately 10% of poly-innervated MNTB neurons along the mouse lifespan, with prevalence in the low frequency region. Moreover, a tonotopy-dependent heterogeneity in CH morphology was observed in young but not in older mice. In conclusion, our data support the notion that age-related hearing impairments can be in part a direct consequence of several structural alterations and circuit remodeling in the brainstem.

A Unified Deep Learning Framework for ssTEM Image Restoration

Serial section transmission electron micro-scopy (ssTEM) reveals biological information at a scale of nanometer and plays an important role in the ultrastructural analysis. However, due to the imperfect preparation of biological samples, ssTEM images are usually degraded with various artifacts that greatly challenge the subsequent analysis and visualization. In this paper, we introduce a unified deep learning framework for ssTEM image restoration which addresses three main types of artifacts, i.e., Support Film Folds (SFF), Staining Precipitates (SP), and Missing Sections (MS). To achieve this goal, we first model the appearance of SFF and SP artifacts by conducting comprehensive analyses on the statistics of real degraded images, relying on which we can then simulate a large number of paired images (degraded/artifacts-free) for training a deep restoration network. Then, we design a coarse-to-fine restoration network consisting of three modules, i.e., interpolation, correction, and fusion. The interpolation module exploits the adjacent artifacts-free images for an initial restoration, while the correction module resorts to the degraded image itself to rectify the artifacts. Finally, the fusion module jointly utilizes the above two results to further improve the restoration fidelity. Experimental results on both synthetic and real test data validate the significantly improved performance of our proposed framework over existing solutions, in terms of both image restoration fidelity and neuron segmentation accuracy. To the best of our knowledge, this is the first unified deep learning framework for ssTEM image restoration from different types of artifacts. Code is available at https://github.com/sydeng99/ssTEM-restoration.

Binary and analog variation of synapses between cortical pyramidal neurons

Learning from experience depends at least in part on changes in neuronal connections. We present the largest map of connectivity to date between cortical neurons of a defined type (layer 2/3 [L2/3] pyramidal cells in mouse primary visual cortex), which was enabled by automated analysis of serial section electron microscopy images with improved handling of image defects (250 × 140 × 90 μm3 volume). We used the map to identify constraints on the learning algorithms employed by the cortex. Previous cortical studies modeled a continuum of synapse sizes by a log-normal distribution. A continuum is consistent with most neural network models of learning, in which synaptic strength is a continuously graded analog variable. Here, we show that synapse size, when restricted to synapses between L2/3 pyramidal cells, is well modeled by the sum of a binary variable and an analog variable drawn from a log-normal distribution. Two synapses sharing the same presynaptic and postsynaptic cells are known to be correlated in size. We show that the binary variables of the two synapses are highly correlated, while the analog variables are not. Binary variation could be the outcome of a Hebbian or other synaptic plasticity rule depending on activity signals that are relatively uniform across neuronal arbors, while analog variation may be dominated by other influences such as spontaneous dynamical fluctuations. We discuss the implications for the longstanding hypothesis that activity-dependent plasticity switches synapses between bistable states.

Deep learning empowered volume delineation of whole-body organs-at-risk for accelerated radiotherapy

In radiotherapy for cancer patients, an indispensable process is to delineate organs-at-risk (OARs) and tumors. However, it is the most time-consuming step as manual delineation is always required from radiation oncologists. Herein, we propose a lightweight deep learning framework for radiotherapy treatment planning (RTP), named RTP-Net, to promote an automatic, rapid, and precise initialization of whole-body OARs and tumors. Briefly, the framework implements a cascade coarse-to-fine segmentation, with adaptive module for both small and large organs, and attention mechanisms for organs and boundaries. Our experiments show three merits: 1) Extensively evaluates on 67 delineation tasks on a large-scale dataset of 28,581 cases; 2) Demonstrates comparable or superior accuracy with an average Dice of 0.95; 3) Achieves near real-time delineation in most tasks with <2 s. This framework could be utilized to accelerate the contouring process in the All-in-One radiotherapy scheme, and thus greatly shorten the turnaround time of patients.

Promoting Connectivity of Network-Like Structures by Enforcing Region Separation

We propose a novel, connectivity-oriented loss function for training deep convolutional networks to reconstruct network-like structures, like roads and irrigation canals, from aerial images. The main idea behind our loss is to express the connectivity of roads, or canals, in terms of disconnections that they create between background regions of the image. In simple terms, a gap in the predicted road causes two background regions, that lie on the opposite sides of a ground truth road, to touch in prediction. Our loss function is designed to prevent such unwanted connections between background regions, and therefore close the gaps in predicted roads. It also prevents predicting false positive roads and canals by penalizing unwarranted disconnections of background regions. In order to capture even short, dead-ending road segments, we evaluate the loss in small image crops. We show, in experiments on two standard road benchmarks and a new data set of irrigation canals, that convnets trained with our loss function recover road connectivity so well that it suffices to skeletonize their output to produce state of the art maps. A distinct advantage of our approach is that the loss can be plugged in to any existing training setup without further modifications.

How innovations in methodology offer new prospects for volume electron microscopy

Detailed knowledge of biological structure has been key in understanding biology at several levels of organisation, from organs to cells and proteins. Volume electron microscopy (volume EM) provides high resolution 3D structural information about tissues on the nanometre scale. However, the throughput rate of conventional electron microscopes has limited the volume size and number of samples that can be imaged. Recent improvements in methodology are currently driving a revolution in volume EM, making possible the structural imaging of whole organs and small organisms. In turn, these recent developments in image acquisition have created or stressed bottlenecks in other parts of the pipeline, like sample preparation, image analysis and data management. While the progress in image analysis is stunning due to the advent of automatic segmentation and server‐based annotation tools, several challenges remain. Here we discuss recent trends in volume EM, emerging methods for increasing throughput and implications for sample preparation, image analysis and data management.

Volume electron microscopy

Life exists in three dimensions, but until the turn of the century most electron microscopy methods provided only 2D image data. Recently, electron microscopy techniques capable of delving deep into the structure of cells and tissues have emerged, collectively called volume electron microscopy (vEM). Developments in vEM have been dubbed a quiet revolution as the field evolved from established transmission and scanning electron microscopy techniques, so early publications largely focused on the bioscience applications rather than the underlying technological breakthroughs. However, with an explosion in the uptake of vEM across the biosciences and fast-paced advances in volume, resolution, throughput and ease of use, it is timely to introduce the field to new audiences. In this Primer, we introduce the different vEM imaging modalities, the specialized sample processing and image analysis pipelines that accompany each modality and the types of information revealed in the data. We showcase key applications in the biosciences where vEM has helped make breakthrough discoveries and consider limitations and future directions. We aim to show new users how vEM can support discovery science in their own research fields and inspire broader uptake of the technology, finally allowing its full adoption into mainstream biological imaging.

H-ProSeg: Hybrid ultrasound prostate segmentation based on explainability-guided mathematical model

BACKGROUND AND OBJECTIVE

Accurate and robust prostate segmentation in transrectal ultrasound (TRUS) images is of great interest for image-guided prostate interventions and prostate cancer diagnosis. However, it remains a challenging task for various reasons, including a missing or ambiguous boundary between the prostate and surrounding tissues, the presence of shadow artifacts, intra-prostate intensity heterogeneity, and anatomical variations.

METHODS

Here, we present a hybrid method for prostate segmentation (H-ProSeg) in TRUS images, using a small number of radiologist-defined seed points as the prior points. This method consists of three subnetworks. The first subnetwork uses an improved principal curve-based model to obtain data sequences consisting of seed points and their corresponding projection index. The second subnetwork uses an improved differential evolution-based artificial neural network for training to decrease the model error. The third subnetwork uses the parameters of the artificial neural network to explain the smooth mathematical description of the prostate contour. The performance of the H-ProSeg method was assessed in 55 brachytherapy patients using Dice similarity coefficient (DSC), Jaccard similarity coefficient (Ω), and accuracy (ACC) values.

RESULTS

The H-ProSeg method achieved excellent segmentation accuracy, with DSC, Ω, and ACC values of 95.8%, 94.3%, and 95.4%, respectively. Meanwhile, the DSC, Ω, and ACC values of the proposed method were as high as 93.3%, 91.9%, and 93%, respectively, due to the influence of Gaussian noise (standard deviation of Gaussian function, σ = 50). Although the σ increased from 10 to 50, the DSC, Ω, and ACC values fluctuated by a maximum of approximately 2.5%, demonstrating the excellent robustness of our method.

CONCLUSIONS

Here, we present a hybrid method for accurate and robust prostate ultrasound image segmentation. The H-ProSeg method achieved superior performance compared with current state-of-the-art techniques. The knowledge of precise boundaries of the prostate is crucial for the conservation of risk structures. The proposed models have the potential to improve prostate cancer diagnosis and therapeutic outcomes.

Candelabrum cells are ubiquitous cerebellar cortex interneurons with specialized circuit properties

To understand how the cerebellar cortex transforms mossy fiber (MF) inputs into Purkinje cell (PC) outputs, it is vital to delineate the elements of this circuit. Candelabrum cells (CCs) are enigmatic interneurons of the cerebellar cortex that have been identified based on their morphology, but their electrophysiological properties, synaptic connections and function remain unknown. Here, we clarify these properties using electrophysiology, single-nucleus RNA sequencing, in situ hybridization and serial electron microscopy in mice. We find that CCs are the most abundant PC layer interneuron. They are GABAergic, molecularly distinct and present in all cerebellar lobules. Their high resistance renders CC firing highly sensitive to synaptic inputs. CCs are excited by MFs and granule cells and are strongly inhibited by PCs. CCs in turn primarily inhibit molecular layer interneurons, which leads to PC disinhibition. Thus, inputs, outputs and local signals converge onto CCs to allow them to assume a unique role in controlling cerebellar output.

Reconstruction of neocortex: Organelles, compartments, cells, circuits, and activity

We assembled a semi-automated reconstruction of L2/3 mouse primary visual cortex from ∼250 × 140 × 90 μm3 of electron microscopic images, including pyramidal and non-pyramidal neurons, astrocytes, microglia, oligodendrocytes and precursors, pericytes, vasculature, nuclei, mitochondria, and synapses. Visual responses of a subset of pyramidal cells are included. The data are publicly available, along with tools for programmatic and three-dimensional interactive access. Brief vignettes illustrate the breadth of potential applications relating structure to function in cortical circuits and neuronal cell biology. Mitochondria and synapse organization are characterized as a function of path length from the soma. Pyramidal connectivity motif frequencies are predicted accurately using a configuration model of random graphs. Pyramidal cells receiving more connections from nearby cells exhibit stronger and more reliable visual responses. Sample code shows data access and analysis.

Partial connectomes of labeled dopaminergic circuits reveal non-synaptic communication and axonal remodeling after exposure to cocaine

Dopaminergic (DA) neurons exert profound influences on behavior including addiction. However, how DA axons communicate with target neurons and how those communications change with drug exposure remains poorly understood. We leverage cell type-specific labeling with large volume serial electron microscopy to detail DA connections in the nucleus accumbens (NAc) of the mouse (Mus musculus) before and after exposure to cocaine. We find that individual DA axons contain different varicosity types based on their vesicle contents. Spatially ordering along individual axons further suggests that varicosity types are non-randomly organized. DA axon varicosities rarely make specific synapses (<2%, 6/410), but instead are more likely to form spinule-like structures (15%, 61/410) with neighboring neurons. Days after a brief exposure to cocaine, DA axons were extensively branched relative to controls, formed blind-ended 'bulbs' filled with mitochondria, and were surrounded by elaborated glia. Finally, mitochondrial lengths increased by ~2.2 times relative to control only in DA axons and NAc spiny dendrites after cocaine exposure. We conclude that DA axonal transmission is unlikely to be mediated via classical synapses in the NAc and that the major locus of anatomical plasticity of DA circuits after exposure to cocaine are large-scale axonal re-arrangements with correlated changes in mitochondria.

Learning and Segmenting Dense Voxel Embeddings for 3D Neuron Reconstruction

We show dense voxel embeddings learned via deep metric learning can be employed to produce a highly accurate segmentation of neurons from 3D electron microscopy images. A "metric graph" on a set of edges between voxels is constructed from the dense voxel embeddings generated by a convolutional network. Partitioning the metric graph with long-range edges as repulsive constraints yields an initial segmentation with high precision, with substantial accuracy gain for very thin objects. The convolutional embedding net is reused without any modification to agglomerate the systematic splits caused by complex "self-contact" motifs. Our proposed method achieves state-of-the-art accuracy on the challenging problem of 3D neuron reconstruction from the brain images acquired by serial section electron microscopy. Our alternative, object-centered representation could be more generally useful for other computational tasks in automated neural circuit reconstruction.

Whole-cell organelle segmentation in volume electron microscopy

Cells contain hundreds of organelles and macromolecular assemblies. Obtaining a complete understanding of their intricate organization requires the nanometre-level, three-dimensional reconstruction of whole cells, which is only feasible with robust and scalable automatic methods. Here, to support the development of such methods, we annotated up to 35 different cellular organelle classes-ranging from endoplasmic reticulum to microtubules to ribosomes-in diverse sample volumes from multiple cell types imaged at a near-isotropic resolution of 4 nm per voxel with focused ion beam scanning electron microscopy (FIB-SEM)¹. We trained deep learning architectures to segment these structures in 4 nm and 8 nm per voxel FIB-SEM volumes, validated their performance and showed that automatic reconstructions can be used to directly quantify previously inaccessible metrics including spatial interactions between cellular components. We also show that such reconstructions can be used to automatically register light and electron microscopy images for correlative studies. We have created an open data and open-source web repository, 'OpenOrganelle', to share the data, computer code and trained models, which will enable scientists everywhere to query and further improve automatic reconstruction of these datasets.

The Mutex Watershed and its Objective: Efficient, Parameter-Free Graph Partitioning

Image partitioning, or segmentation without semantics, is the task of decomposing an image into distinct segments, or equivalently to detect closed contours. Most prior work either requires seeds, one per segment; or a threshold; or formulates the task as multicut / correlation clustering, an NP-hard problem. Here, we propose an efficient algorithm for graph partitioning, the "Mutex Watershed". Unlike seeded watershed, the algorithm can accommodate not only attractive but also repulsive cues, allowing it to find a previously unspecified number of segments without the need for explicit seeds or a tunable threshold. We also prove that this simple algorithm solves to global optimality an objective function that is intimately related to the multicut / correlation clustering integer linear programming formulation. The algorithm is deterministic, very simple to implement, and has empirically linearithmic complexity. When presented with short-range attractive and long-range repulsive cues from a deep neural network, the Mutex Watershed gives the best results currently known for the competitive ISBI 2012 EM segmentation benchmark.

Whole-body integration of gene expression and single-cell morphology

Animal bodies are composed of cell types with unique expression programs that implement their distinct locations, shapes, structures, and functions. Based on these properties, cell types assemble into specific tissues and organs. To systematically explore the link between cell-type-specific gene expression and morphology, we registered an expression atlas to a whole-body electron microscopy volume of the nereid Platynereis dumerilii. Automated segmentation of cells and nuclei identifies major cell classes and establishes a link between gene activation, chromatin topography, and nuclear size. Clustering of segmented cells according to gene expression reveals spatially coherent tissues. In the brain, genetically defined groups of neurons match ganglionic nuclei with coherent projections. Besides interneurons, we uncover sensory-neurosecretory cells in the nereid mushroom bodies, which thus qualify as sensory organs. They furthermore resemble the vertebrate telencephalon by molecular anatomy. We provide an integrated browser as a Fiji plugin for remote exploration of all available multimodal datasets.

•

A cellular atlas integrates gene expression and ultrastructure for an entire annelid

•

Morphometry of all segmented cells, nuclei, and chromatin categorizes cell classes

•

Molecular anatomy and projectome of head ganglionic nuclei and mushroom bodies

•

An open-source browser for multimodal big image data exploration and analysis

Automatic detection of synaptic partners in a whole-brain Drosophila electron microscopy data set

We develop an automatic method for synaptic partner identification in insect brains and use it to predict synaptic partners in a whole-brain electron microscopy dataset of the fruit fly. The predictions can be used to infer a connectivity graph with high accuracy, thus allowing fast identification of neural pathways. To facilitate circuit reconstruction using our results, we develop CIRCUITMAP, a user interface add-on for the circuit annotation tool CATMAID.

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.

Weakly supervised serous retinal detachment segmentation in SD-OCT images by two-stage learning

Automated lesion segmentation is one of the important tasks for the quantitative assessment of retinal diseases in SD-OCT images. Recently, deep convolutional neural networks (CNN) have shown promising advancements in the field of automated image segmentation, whereas they always benefit from large-scale datasets with high-quality pixel-wise annotations. Unfortunately, obtaining accurate annotations is expensive in both human effort and finance. In this paper, we propose a weakly supervised two-stage learning architecture to detect and further segment central serous chorioretinopathy (CSC) retinal detachment with only image-level annotations. Specifically, in the first stage, a Located-CNN is designed to detect the location of lesion regions in the whole SD-OCT retinal images, and highlight the distinguishing regions. To generate available a pseudo pixel-level label, the conventional level set method is employed to refine the distinguishing regions. In the second stage, we customize the active-contour loss function in deep networks to achieve the effective segmentation of the lesion area. A challenging dataset is used to evaluate our proposed method, and the results demonstrate that the proposed method consistently outperforms some current models trained with a different level of supervision, and is even as competitive as those relying on stronger supervision. To our best knowledge, we are the first to achieve CSC segmentation in SD-OCT images using weakly supervised learning, which can greatly reduce the labeling efforts.

HIVE-Net: Centerline-aware hierarchical view-ensemble convolutional network for mitochondria segmentation in EM images

BACKGROUND AND OBJECTIVE

With the advancement of electron microscopy (EM) imaging technology, neuroscientists can investigate the function of various intracellular organelles, e.g, mitochondria, at nano-scale. Semantic segmentation of electron microscopy (EM) is an essential step to efficiently obtain reliable morphological statistics. Despite the great success achieved using deep convolutional neural networks (CNNs), they still produce coarse segmentations with lots of discontinuities and false positives for mitochondria segmentation.

METHODS

In this study, we introduce a centerline-aware multitask network by utilizing centerline as an intrinsic shape cue of mitochondria to regularize the segmentation. Since the application of 3D CNNs on large medical volumes is usually hindered by their substantial computational cost and storage overhead, we introduce a novel hierarchical view-ensemble convolution (HVEC), a simple alternative of 3D convolution to learn 3D spatial contexts using more efficient 2D convolutions. The HVEC enables both decomposing and sharing multi-view information, leading to increased learning capacity.

RESULTS

Extensive validation results on two challenging benchmarks show that, the proposed method performs favorably against the state-of-the-art methods in accuracy and visual quality but with a greatly reduced model size. Moreover, the proposed model also shows significantly improved generalization ability, especially when training with quite limited amount of training data. Detailed sensitivity analysis and ablation study have also been conducted, which show the robustness of the proposed model and effectiveness of the proposed modules.

CONCLUSIONS

The experiments highlighted that the proposed architecture enables both simplicity and efficiency leading to increased capacity of learning spatial contexts. Moreover, incorporating shape cues such as centerline information is a promising approach to improve the performance of mitochondria segmentation.

DeepACSON automated segmentation of white matter in 3D electron microscopy

Tracing the entirety of ultrastructures in large three-dimensional electron microscopy (3D-EM) images of the brain tissue requires automated segmentation techniques. Current segmentation techniques use deep convolutional neural networks (DCNNs) and rely on high-contrast cellular membranes and high-resolution EM volumes. On the other hand, segmenting low-resolution, large EM volumes requires methods to account for severe membrane discontinuities inescapable. Therefore, we developed DeepACSON, which performs DCNN-based semantic segmentation and shape-decomposition-based instance segmentation. DeepACSON instance segmentation uses the tubularity of myelinated axons and decomposes under-segmented myelinated axons into their constituent axons. We applied DeepACSON to ten EM volumes of rats after sham-operation or traumatic brain injury, segmenting hundreds of thousands of long-span myelinated axons, thousands of cell nuclei, and millions of mitochondria with excellent evaluation scores. DeepACSON quantified the morphology and spatial aspects of white matter ultrastructures, capturing nanoscopic morphological alterations five months after the injury.

Dense neuronal reconstruction through X-ray holographic nano-tomography

Imaging neuronal networks provides a foundation for understanding the nervous system, but resolving dense nanometer-scale structures over large volumes remains challenging for light microscopy (LM) and electron microscopy (EM). Here we show that X-ray holographic nano-tomography (XNH) can image millimeter-scale volumes with sub-100-nm resolution, enabling reconstruction of dense wiring in Drosophila melanogaster and mouse nervous tissue. We performed correlative XNH and EM to reconstruct hundreds of cortical pyramidal cells and show that more superficial cells receive stronger synaptic inhibition on their apical dendrites. By combining multiple XNH scans, we imaged an adult Drosophila leg with sufficient resolution to comprehensively catalog mechanosensory neurons and trace individual motor axons from muscles to the central nervous system. To accelerate neuronal reconstructions, we trained a convolutional neural network to automatically segment neurons from XNH volumes. Thus, XNH bridges a key gap between LM and EM, providing a new avenue for neural circuit discovery.

MitoEM Dataset: Large-scale 3D Mitochondria Instance Segmentation from EM Images

Electron microscopy (EM) allows the identification of intracellular organelles such as mitochondria, providing insights for clinical and scientific studies. However, public mitochondria segmentation datasets only contain hundreds of instances with simple shapes. It is unclear if existing methods achieving human-level accuracy on these small datasets are robust in practice. To this end, we introduce the MitoEM dataset, a 3D mitochondria instance segmentation dataset with two (30μm)3 volumes from human and rat cortices respectively, 3, 600× larger than previous benchmarks. With around 40K instances, we find a great diversity of mitochondria in terms of shape and density. For evaluation, we tailor the implementation of the average precision (AP) metric for 3D data with a 45× speedup. On MitoEM, we find existing instance segmentation methods often fail to correctly segment mitochondria with complex shapes or close contacts with other instances. Thus, our MitoEM dataset poses new challenges to the field. We release our code and data: https://donglaiw.github.io/page/mitoEM/index.html.

Multimodal Medical Image Fusion using Rolling Guidance Filter with CNN and Nuclear Norm Minimization

BACKGROUND

Medical image fusion is very important for the diagnosis and treatment of diseases. In recent years, there have been a number of different multi-modal medical image fusion algorithms that can provide delicate contexts for disease diagnosis more clearly and more conveniently. Recently, nuclear norm minimization and deep learning have been used effectively in image processing.

METHODS

A multi-modality medical image fusion method using a rolling guidance filter (RGF) with a convolutional neural network (CNN) based feature mapping and nuclear norm minimization (NNM) is proposed. At first, we decompose medical images to base layer components and detail layer components by using RGF. In the next step, we get the basic fused image through the pretrained CNN model. The CNN model with pre-training is used to obtain the significant characteristics of the base layer components. And we can compute the activity level measurement from the regional energy of CNN-based fusion maps. Then, a detail fused image is gained by NNM. That is, we use NNM to fuse the detail layer components. At last, the basic and detail fused images are integrated into the fused result.

RESULTS

From the comparison with the most advanced fusion algorithms, the results of experiments indicate that this fusion algorithm has the best effect in visual evaluation and objective standard.

CONCLUSION

The fusion algorithm using RGF and CNN-based feature mapping, combined with NNM, can improve fusion effects and suppress artifacts and blocking effects in the fused results.

Accurate and versatile 3D segmentation of plant tissues at cellular resolution

Quantitative analysis of plant and animal morphogenesis requires accurate segmentation of individual cells in volumetric images of growing organs. In the last years, deep learning has provided robust automated algorithms that approach human performance, with applications to bio-image analysis now starting to emerge. Here, we present PlantSeg, a pipeline for volumetric segmentation of plant tissues into cells. PlantSeg employs a convolutional neural network to predict cell boundaries and graph partitioning to segment cells based on the neural network predictions. PlantSeg was trained on fixed and live plant organs imaged with confocal and light sheet microscopes. PlantSeg delivers accurate results and generalizes well across different tissues, scales, acquisition settings even on non plant samples. We present results of PlantSeg applications in diverse developmental contexts. PlantSeg is free and open-source, with both a command line and a user-friendly graphical interface.

Computer Vision and Deep Learning Techniques for the Analysis of Drone-Acquired Forest Images, a Transfer Learning Study

Unmanned Aerial Vehicles (UAV) are becoming an essential tool for evaluating the status and the changes in forest ecosystems. This is especially important in Japan due to the sheer magnitude and complexity of the forest area, made up mostly of natural mixed broadleaf deciduous forests. Additionally, Deep Learning (DL) is becoming more popular for forestry applications because it allows for the inclusion of expert human knowledge into the automatic image processing pipeline. In this paper we study and quantify issues related to the use of DL with our own UAV-acquired images in forestry applications such as: the effect of Transfer Learning (TL) and the Deep Learning architecture chosen or whether a simple patch-based framework may produce results in different practical problems. We use two different Deep Learning architectures (ResNet50 and UNet), two in-house datasets (winter and coastal forest) and focus on two separate problem formalizations (Multi-Label Patch or MLP classification and semantic segmentation). Our results show that Transfer Learning is necessary to obtain satisfactory outcome in the problem of MLP classification of deciduous vs evergreen trees in the winter orthomosaic dataset (with a 9.78% improvement from no transfer learning to transfer learning from a a general-purpose dataset). We also observe a further 2.7% improvement when Transfer Learning is performed from a dataset that is closer to our type of images. Finally, we demonstrate the applicability of the patch-based framework with the ResNet50 architecture in a different and complex example: Detection of the invasive broadleaf deciduous black locust (Robinia pseudoacacia) in an evergreen coniferous black pine (Pinus thunbergii) coastal forest typical of Japan. In this case we detect images containing the invasive species with a 75% of True Positives (TP) and 9% False Positives (FP) while the detection of native trees was 95% TP and 10% FP.

The natverse, a versatile toolbox for combining and analysing neuroanatomical data

To analyse neuron data at scale, neuroscientists expend substantial effort reading documentation, installing dependencies and moving between analysis and visualisation environments. To facilitate this, we have developed a suite of interoperable open-source R packages called the natverse. The natverse allows users to read local and remote data, perform popular analyses including visualisation and clustering and graph-theoretic analysis of neuronal branching. Unlike most tools, the natverse enables comparison across many neurons of morphology and connectivity after imaging or co-registration within a common template space. The natverse also enables transformations between different template spaces and imaging modalities. We demonstrate tools that integrate the vast majority of Drosophila neuroanatomical light microscopy and electron microscopy connectomic datasets. The natverse is an easy-to-use environment for neuroscientists to solve complex, large-scale analysis challenges as well as an open platform to create new code and packages to share with the community.

Generative and discriminative model-based approaches to microscopic image restoration and segmentation

Image processing is one of the most important applications of recent machine learning (ML) technologies. Convolutional neural networks (CNNs), a popular deep learning-based ML architecture, have been developed for image processing applications. However, the application of ML to microscopic images is limited as microscopic images are often 3D/4D, that is, the image sizes can be very large, and the images may suffer from serious noise generated due to optics. In this review, three types of feature reconstruction applications to microscopic images are discussed, which fully utilize the recent advancements in ML technologies. First, multi-frame super-resolution is introduced, based on the formulation of statistical generative model-based techniques such as Bayesian inference. Second, data-driven image restoration is introduced, based on supervised discriminative model-based ML technique. In this application, CNNs are demonstrated to exhibit preferable restoration performance. Third, image segmentation based on data-driven CNNs is introduced. Image segmentation has become immensely popular in object segmentation based on electron microscopy (EM); therefore, we focus on EM image processing.

Tracing in 2D to reduce the annotation effort for 3D deep delineation of linear structures

The difficulty of obtaining annotations to build training databases still slows down the adoption of recent deep learning approaches for biomedical image analysis. In this paper, we show that we can train a Deep Net to perform 3D volumetric delineation given only 2D annotations in Maximum Intensity Projections (MIP) of the training volumes. This significantly reduces the annotation time: We conducted a user study that suggests that annotating 2D projections is on average twice as fast as annotating the original 3D volumes. Our technical contribution is a loss function that evaluates a 3D prediction against annotations of 2D projections. It is inspired by space carving, a classical approach to reconstructing complex 3D shapes from arbitrarily-positioned cameras. It can be used to train any deep network with volumetric output, without the need to change the network's architecture. Substituting the loss is all it takes to enable 2D annotations in an existing training setup. In extensive experiments on 3D light microscopy images of neurons and retinal blood vessels, and on Magnetic Resonance Angiography (MRA) brain scans, we show that, when trained on projection annotations, deep delineation networks perform as well as when they are trained using costlier 3D annotations.

Learning cellular morphology with neural networks

Reconstruction and annotation of volume electron microscopy data sets of brain tissue is challenging but can reveal invaluable information about neuronal circuits. Significant progress has recently been made in automated neuron reconstruction as well as automated detection of synapses. However, methods for automating the morphological analysis of nanometer-resolution reconstructions are less established, despite the diversity of possible applications. Here, we introduce cellular morphology neural networks (CMNs), based on multi-view projections sampled from automatically reconstructed cellular fragments of arbitrary size and shape. Using unsupervised training, we infer morphology embeddings (Neuron2vec) of neuron reconstructions and train CMNs to identify glia cells in a supervised classification paradigm, which are then used to resolve neuron reconstruction errors. Finally, we demonstrate that CMNs can be used to identify subcellular compartments and the cell types of neuron reconstructions.