High-throughput segmentation of unmyelinated axons by deep learning

A method for quantitative spatial analysis of immunolabeled fibers at regenerative electrode interfaces

BACKGROUND

Regenerative electrodes are being explored as robust peripheral nerve interfaces for neuro-prosthetic control and sensory feedback. Current designs differ in electrode number, spatial arrangement, and porosity which impacts the regeneration, activation, and spatial distribution of fibers at the device interface. Knowledge of sensory and motor fiber distributions are important in optimizing selective fiber activation and recording.

NEW METHOD

We use confocal microscopy and immunofluorescence methods to conduct spatial analysis of immunolabeled fibers across whole nerve cross sections.

RESULTS

This protocol was implemented to characterize motor fiber distribution within 3 macro-sieve electrode regenerated (MSE), 3 silicone-conduit regenerated, and 3 unmanipulated control rodent sciatic nerves. Total motor fiber counts were 1485 [SD: +/- 50.11], 1899 [SD: +/- 359], and 5732 [SD: +/- 1410] for control, MSE, and conduit nerves respectively. MSE motor fiber distributions exhibited evidence of deviation from complete spatial randomness and evidence of dispersion and clustering tendencies at varying scales. Notably, MSE motor fibers exhibited clustering within the central portion of the cross section, whereas conduit regenerated motor fibers exhibited clustering along the periphery.

COMPARISON WITH EXISTING METHODS

Prior exploration of fiber distributions at regenerative interfaces was limited to either quadrant-based density analysis of randomly sampled subregions or qualitative description. This method extends existing sample preparation and microscopy techniques to quantitatively assess immunolabeled fiber distributions within whole nerve cross-sections.

CONCLUSIONS

This approach is an effective way to examine the spatial organization of fiber subsets at regenerative electrode interfaces, enabling robust assessment of fiber distributions relative to electrode arrangement.

Nonbinary 2D Distribution Tool Maps Autonomic Nerve Fiber Clustering in Lumbosacral Ventral Roots of Rhesus Macaques

Neuromodulation of the peripheral nervous system (PNS) by electrical stimulation may augment autonomic function after injury or in neurodegenerative disorders. Nerve fiber size, myelination, and distance between individual fibers and the stimulation electrode may influence response thresholds to electrical stimulation. However, information on the spatial distribution of nerve fibers within the PNS is sparse. We developed a new two-dimensional (2D) morphological mapping tool to assess spatial heterogeneity and clustering of nerve fibers. The L6-S3 ventral roots (VRs) in rhesus macaques were used as a model system to map preganglionic parasympathetic, γ-motor, and α-motor fibers. Random and ground truth distributions of nerve fiber centroids were determined for each VR by light microscopy. The proposed tool allows for nonbinary determinations of fiber heterogeneity by defining the minimum distance between nerve fibers for cluster inclusion and comparisons with random fiber distributions for each VR. There was extensive variability in the relative composition of nerve fiber types and degree of 2D fiber heterogeneity between different L6-S3 VR levels within and across different animals. There was a positive correlation between the proportion of autonomic fibers and the degree of nerve fiber clustering. Nerve fiber cluster heterogeneity between VRs may contribute to varied functional outcomes from neuromodulation.

AxoDetect: an automated nerve image segmentation and quantification workflow for computational nerve modeling

Objective.Bioelectronic treatments targeting near-organ innervation have unprecedented clinical applications. Particularly in the spleen, the inhibition of the cholinergic inflammatory response by near-organ nerve stimulation has potential to replace pharmacological treatments in chronic and autoimmune diseases. A caveat is that the optimization of therapeutic stimulation parameters relies onin vivoexperimentation, which becomes challenging due to the small nerve diameters (2 μm), complex anatomy, and mixed axon type composition of the autonomic nerves. Effective development ofin silicomodels requires tools which allow for fast and efficient quantification of axonal composition of specific nerves. Current approaches to generate such information rely on manual image segmentation and quantification.Approach.We developed a combined image-segmentation and model-generation software called AxoDetect: a target- and format-agnostic computer vision algorithm which can segment myelin, endo/epineurium, and both myelinated and unmyelinated fibers from a nerve image without training.Main results.AxoDetect is over 10 times faster on average when compared with current automatic methods while maintaining flexibility through the use of tunable pixel threshold filters to detect different types of tissue. When compared to a distribution-based and a manually segmented model of the splenic nerve terminal branch 1, the model generated with AxoDetect had comparable threshold prediction and was able to accurately detect an increase in activation threshold caused by the addition of surrounding fat tissue to the modeled nerve.Significance.AxoDetect contributes to the acceleration of neuromodulation treatment development through faster model design and iteration without requiring training. Furthermore, the computer vision approach and tunable nature of the filters in our method allow for its use in a variety of histological applications. Our approach will impact not only the study of nerves but also the design of implantable neural interfaces to enhance bioelectronic therapeutic options.

Computational models of compound nerve action potentials: Efficient filter-based methods to quantify effects of tissue conductivities, conduction distance, and nerve fiber parameters

BACKGROUND

Peripheral nerve recordings can enhance the efficacy of neurostimulation therapies by providing a feedback signal to adjust stimulation settings for greater efficacy or reduced side effects. Computational models can accelerate the development of interfaces with high signal-to-noise ratio and selective recording. However, validation and tuning of model outputs against in vivo recordings remains computationally prohibitive due to the large number of fibers in a nerve.

METHODS

We designed and implemented highly efficient modeling methods for simulating electrically evoked compound nerve action potential (CNAP) signals. The method simulated a subset of fiber diameters present in the nerve using NEURON, interpolated action potential templates across fiber diameters, and filtered the templates with a weighting function derived from fiber-specific conduction velocity and electromagnetic reciprocity outputs of a volume conductor model. We applied the methods to simulate CNAPs from rat cervical vagus nerve.

RESULTS

Brute force simulation of a rat vagal CNAP with all 1,759 myelinated and 13,283 unmyelinated fibers in NEURON required 286 and 15,860 CPU hours, respectively, while filtering interpolated templates required 30 and 38 seconds on a desktop computer while maintaining accuracy. Modeled CNAP amplitude could vary by over two orders of magnitude depending on tissue conductivities and cuff opening within experimentally relevant ranges. Conduction distance and fiber diameter distribution also strongly influenced the modeled CNAP amplitude, shape, and latency. Modeled and in vivo signals had comparable shape, amplitude, and latency for myelinated fibers but not for unmyelinated fibers.

CONCLUSIONS

Highly efficient methods of modeling neural recordings quantified the large impact that tissue properties, conduction distance, and nerve fiber parameters have on CNAPs. These methods expand the computational accessibility of neural recording models, enable efficient model tuning for validation, and facilitate the design of novel recording interfaces for neurostimulation feedback and understanding physiological systems.

A flexible, thin-film microchannel electrode array device for selective subdiaphragmatic vagus nerve recording

The vagus nerve (VN) plays an important role in regulating physiological conditions in the gastrointestinal (GI) tract by communicating via the parasympathetic pathway to the enteric nervous system (ENS). However, the lack of knowledge in the neurophysiology of the VN and GI tract limits the development of advanced treatments for autonomic dysfunctions related to the VN. To better understand the complicated underlying mechanisms of the VN-GI tract neurophysiology, it is necessary to use an advanced device enabled by microfabrication technologies. Among several candidates including intraneural probe array and extraneural cuff electrodes, microchannel electrode array devices can be used to interface with smaller numbers of nerve fibers by securing them in the separate channel structures. Previous microchannel electrode array devices to interface teased nerve structures are relatively bulky with thickness around 200 µm. The thick design can potentially harm the delicate tissue structures, including the nerve itself. In this paper, we present a flexible thin film based microchannel electrode array device (thickness: 11.5 µm) that can interface with one of the subdiaphragmatic nerve branches of the VN in a rat. We demonstrated recording evoked compound action potentials (ECAP) from a transected nerve ending that has multiple nerve fibers. Moreover, our analysis confirmed that the signals are from C-fibers that are critical in regulating autonomic neurophysiology in the GI tract.

High-throughput image analysis with deep learning captures heterogeneity and spatial relationships after kidney injury

Recovery from acute kidney injury can vary widely in patients and in animal models. Immunofluorescence staining can provide spatial information about heterogeneous injury responses, but often only a fraction of stained tissue is analyzed. Deep learning can expand analysis to larger areas and sample numbers by substituting for time-intensive manual or semi-automated quantification techniques. Here we report one approach to leverage deep learning tools to quantify heterogenous responses to kidney injury that can be deployed without specialized equipment or programming expertise. We first demonstrated that deep learning models generated from small training sets accurately identified a range of stains and structures with performance similar to that of trained human observers. We then showed this approach accurately tracks the evolution of folic acid induced kidney injury in mice and highlights spatially clustered tubules that fail to repair. We then demonstrated that this approach captures the variation in recovery across a robust sample of kidneys after ischemic injury. Finally, we showed markers of failed repair after ischemic injury were correlated both spatially within and between animals and that failed repair was inversely correlated with peritubular capillary density. Combined, we demonstrate the utility and versatility of our approach to capture spatially heterogenous responses to kidney injury.

A novel statistical methodology for quantifying the spatial arrangements of axons in peripheral nerves

A thorough understanding of the neuroanatomy of peripheral nerves is required for a better insight into their function and the development of neuromodulation tools and strategies. In biophysical modeling, it is commonly assumed that the complex spatial arrangement of myelinated and unmyelinated axons in peripheral nerves is random, however, in reality the axonal organization is inhomogeneous and anisotropic. Present quantitative neuroanatomy methods analyze peripheral nerves in terms of the number of axons and the morphometric characteristics of the axons, such as area and diameter. In this study, we employed spatial statistics and point process models to describe the spatial arrangement of axons and Sinkhorn distances to compute the similarities between these arrangements (in terms of first- and second-order statistics) in various vagus and pelvic nerve cross-sections. We utilized high-resolution transmission electron microscopy (TEM) images that have been segmented using a custom-built high-throughput deep learning system based on a highly modified U-Net architecture. Our findings show a novel and innovative approach to quantifying similarities between spatial point patterns using metrics derived from the solution to the optimal transport problem. We also present a generalizable pipeline for quantitative analysis of peripheral nerve architecture. Our data demonstrate differences between male- and female-originating samples and similarities between the pelvic and abdominal vagus nerves.