Smart engineering solutions to assist the design and optimization of miniaturized neural interfaces that provide targeted and selective stimulation of autonomic and peripheral nerves are critical for the advancement of bioelectronic medicine and for the optimization of neural prostheses. A major challenge is the identification of robust stimulation protocols despite the high variability of nerve anatomies between subject and species, which leads to considerable uncertainty in treatment doses and often to the failure of clinical trials and translational studies.Image‐based modelling of complex neurostimulation scenarios with hybrid electromagnetic and neuronal (EM‐neuronal) simulations and detailed implant models that feature neuroelectric models of nerves derived from medical/histological images, is a promising method to address this issue. We developed a computational pipeline in the simulation platform Sim4Life (ZMT Zurich MedTech AG, Switzerland) and as a dedicated service within the open‐source, web‐based O2SPARC platform developed by the IT’IS Foundation within the NIH SPARC Initiative. The pipeline allows semi‐automatic segmentation of nerve cross section images, fine‐grained tissue identification and generation of detailed 3D nerve models to accelerate the creation of complex neurostimulation nerve models. Major tissues, such as endoneurium, fat, connective tissue and epineurium, are identified by different computer vision approaches.Specifically, a number of tools have been developed to facilitate and automatize the meshing of labelled nerve segmentations, to extrude these meshes along nerve trajectories within computational human anatomical phantoms, and to enable the execution of EM simulations with new high performance low‐frequency unstructured EM solvers. These permit the modelling of tissues with spatially varying anisotropy (e.g., within the fascicles) as well as thin insulating layers such as the perineurium. Neuronal simulations are performed with parametrized electrophysiological models of myelinated sensory or motor axons, or unmyelinated axons. Stimulation thresholds can be automatically determined to predict recruitment curves and selectivity indices to optimize spatial selectivity of stimulation or the design of interfaces.To illustrate the benefits of our newly developed nerve segmentation pipeline for bioelectronic medicine, we provide examples of segmentation of multiple nerve cross‐sections from online databases. Pipeline performances and the different strategies are reported using a semantic (reliability of tissue segmentation) and a functional (reliability of prediction of stimulation thresholds) metric.The developed pipeline is ready to be used for different applications, such as the assessment of implant performances, the investigation of safety‐related issues (e.g. broken leads, acute vs chronic stimulation), and the planning of translational studies or the prediction of experimental outcomes.Support or Funding InformationNIH SPARC Initiative 1OT3OD025348‐01Example of developed pipeline, from a semi‐automatic segmentation of a nerve cross section to model prediction passing through 2D and 3D multidomain meshing and hybrid EM neuronal simulations.Figure 1
Read full abstract