Information and Communication Theoretical Understanding and Treatment of Spinal Cord Injuries: State-of-the-art and Research Challenges
Ozgur B. Akan, Hamideh Ramezani, Meltem Civas, Oktay Cetinkaya, Bilgesu A. Bilgin, Naveed A. Abbasi
IInformation and Communication TheoreticalUnderstanding and Treatment of Spinal CordInjuries: State-of-the-art and Research Challenges
Ozgur B. Akan,
Fellow, IEEE , Hamideh Ramezani,
Student Member, IEEE , Meltem Civas,
Student Member, IEEE ,Oktay Cetinkaya,
Member, IEEE , Bilgesu A. Bilgin,
Member, IEEE and Naveed A. Abbasi,
Student Member, IEEE , Abstract —Among the various key networks in the human body,the nervous system occupies central importance. The debilitatingeffects of spinal cord injuries (SCI) impact a significant numberof people throughout the world, and to date, there is no satisfac-tory method to treat them. In this paper, we review the majortreatment techniques for SCI that include promising solutionsbased on information and communication technology (ICT)and identify the key characteristics of such systems. We thenintroduce two novel ICT-based treatment approaches for SCI.The first proposal is based on neural interface systems (NIS) withenhanced feedback, where the external machines are interfacedwith the brain and the spinal cord such that the brain signals aredirectly routed to the limbs for movement. The second proposalrelates to the design of self-organizing artificial neurons (ANs)that can be used to replace the injured or dead biological neurons.Apart from SCI treatment, the proposed methods may also beutilized as enabling technologies for neural interface applicationsby acting as bio-cyber interfaces between the nervous system andmachines. Furthermore, under the framework of Internet of Bio-Nano Things (IoBNT), experience gained from SCI treatmenttechniques can be transferred to nano communication researchto develop intra-body applications, including remote monitoringof the body and automatized targeted drug delivery.
Index Terms —Spinal cord injuries, spinal treatments, neuralinterface systems, artificial neurons.
I. I
NTRODUCTION
The nervous system is one of the most important networksof the body that forms the basis of human intellect and storesour experiences as memories. It is composed of networksof neurons connected through synapses to receive, transmit,and process information in the body. The manipulation ofinformation by the nervous system is still not entirely under-stood. Thus, an information and communication technology
O. B. Akan is with Next-generation and Wireless Communications Labo-ratory, College of Engineering, Koc University, Istanbul, Turkey and Internetof Everything Group, Department of Engineering, University of Cambridge,UK. (e-mail: [email protected].)H. Ramezani and B. A. Bilgin are with the Internet of Everything (IoE)Group, Electrical Engineering Division, Department of Engineering, Univer-sity of Cambridge. (e-mails: { hr404, bab46 } @cam.ac.uk.)M. Civas, O. Cetinkaya, and N. A. Abbasi are with the Next-generationand Wireless Communications Laboratory (NWCL), Department of Electri-cal and Electronics Engineering, Koc University, Istanbul, Turkey (e-mails: { mcivas16, okcetinkaya13, nabbasi13 } @ku.edu.tr).This work was supported in part by ERC project MINERVA (ERC-2013-CoG TABLE I: Types of SCI and Corresponding Percentages.
SCI Types Percentage
Incomplete Tetraplegia
Incomplete Paraplegia . Complete Paraplegia
Complete Tetraplegia . Recovery . (ICT) framework is required to evaluate the relationships ofinformation exchange between various entities of the nervoussystem, and then to address the associated problems [1].Communication between the brain and limbs is maintainedby the spinal cord, which is comprised of ascending anddescending spinal pathways [2]. Each spinal pathway consti-tutes of neurons having long axons, projecting between thebrain and spinal cord. In patients with spinal cord injuries(SCI), advanced amyotrophic lateral sclerosis (ALS), or brain-stem strokes, the brain is functional; however, its signalsare not transmitted to the muscles due to interruption in thetransmission pathways.The social and economic burden of SCI is globally exten-sive. According to the findings [3], 0.93 million new casesof SCI occurred in 2016, while prevalent cases were 27.04million. SCI results in partial loss of motor and/or sensoryfunctions in the incomplete injury. In the case of completeSCI, the patient suffers from the total loss of motor and sensoryabilities. Statistics for the United States provided in Table Ireveal that recovery after injury is very unlikely, and in mostcases SCI causes neurological deficits in all limbs and torso,i.e., tetraplegia [4]. In the less severe form of SCI, paraplegia,the arms are not affected.Two main approaches exist for re-establishing the brain-body connection of the patients suffering from SCI, which wegroup as, biological methods and ICT-based treatment tech-niques. The latter is the main interest of this paper since it hasadvantageous aspects in terms of functional outcome, humantrials, and safety. The ICT-based treatment techniques of SCIcan be achieved by (i) the use of neural interface systems(NIS), and (ii) deployment of artificial neurons (ANs). NISwere first developed for animals [5]–[10] and then utilized for a r X i v : . [ q - b i o . N C ] M a r uman patients to extract control signals from nervous systemand operate assistive devices, including computer cursors [11]–[15] and robotic prostheses [16]–[19]. More importantly, theywere also used to apply electrical stimulations to the nervoussystem or muscles for restoring motor capabilities [20]–[25],which can induce partial neurological recovery [26], [27]. Thedeployment of ANs, on the other hand, targets to replaceinjured or dead biological neurons. This is a relatively recentdirection, and there is still no major work that can qualifyitself as a fully-functional AN.In this paper, we first discuss the mechanisms of SCI to givesome insight on these injuries. We then present the existingtreatment techniques including biological, NIS, and AN-basedapproaches. Specifically, for the NIS and the AN, where theICT techniques may benefit from the existing designs, we alsoidentify the key characteristics of a successful treatment tech-nique. Based on the preceding findings and discussions, wethen present two potential solutions that may be able to treatSCI with their respective methodologies. The first approach isbased on NIS, where we identify key areas for improvementin the existing setups. We point out important areas, suchas feedback, that are yet unexplored. The second proposalis for the development of ANs based on nanomachines thatact as a bridge to deliver neural signals across the injury site.The proposed AN will incorporate plasticity along with self-organization, and will thus be able to adapt in the injury areato make new nervous connections. Additionally, the machinesare proposed to include energy harvesting (EH) so that theyare truly independent in nature. Apart from SCI treatment,the proposed schemes can be used as bio-cyber interfacesin a variety of applications and also motivate for Internetof Bio-Nano Things (IoBNT) and intra-body nanonetworksapplications.The rest of this paper is organized as follows. In Sec-tion II, we present an overview of nervous communication, themechanisms for SCI, and some key treatment approaches. Theexisting ICT-based treatment techniques for SCI are detailedin Section III. Section IV introduces our proposed methods,whereas Section V provides some future directions in SCItreatment as well as other potential applications for the SCItreatment techniques. Finally, Section VI concludes the paper.II. S PINAL C ORD I NJURY T REATMENT T ECHNIQUES
In this section, we present an overview of nervous systemcommunication, how it is disrupted by SCI during variousphases of the injuries, the existing biological treatment tech-niques for SCI, and the key differences between the biologicaland ICT-based treatment techniques.
A. Neuro-spike Communication
Communication in the ultra large-scale nervous network,known as neuro-spike communication, takes place throughelectrical or chemical synapses with the latter occurring morefrequently [28]. Information encoded in electrical impulsesis transmitted to the post-synaptic neuron using three steps. (i) Axonal transmission : the electrical impulses pass through
CystMyelin Glial ScarDamaged Axons Axon
Fig. 1: SCI and the formation of glial scars. the axon of the pre-synaptic neuron until they reach the pre-synaptic terminals. (ii) Synaptic propagation : arrival of anelectrical impulse to the pre-synaptic terminal initiates therelease of neurotransmitter into the synaptic cleft, i.e., the gapbetween input and output neuron. The neurotransmitters thendiffuse though the synaptic cleft to reach the post-synapticneuron and bind to the receptors located on its membrane.This changes the post-synaptic membrane potential. (iii) Spikegeneration : if the total change in the post-synaptic membranepotential is greater than the spiking threshold, the post-synapticneuron fires an electrical impulse.Several studies exist in the literature on modeling and an-alyzing the performance of neuro-spike communication [28]–[36] and the nervous nanonetwork [37], [38] consideringthe functionality of healthy neurons. Any disruption in thefunctionality of neurons, which can result from diseases orinjuries, adversely affects the performance of this nanoscalecommunication network.
B. Injury Mechanisms
Neurons inflicted by spinal trauma experience the effectsof an injury in three consecutive phases, namely, primary,secondary, and chronic injuries described below. • Primary injury:
This refers to the condition within minutesafter the injury happens. The injury immediately disruptsaxonal pathways and damages cell membranes depending onthe physical trauma [39]. The ischaemia, i.e., toxic chemicalrelease from damaged membranes, initiates the secondaryinjury [39]. Many active neural connections, and thus neuro-spike communications, are affected by this extracellulartoxicity. • Secondary injury:
Following hours to weeks after the injury,SCI develops a condition called secondary injury [40]. Inthis phase, deteriorated toxic environment in the spinal cordcauses death of neurons as well as glial cells. [41]. Glialcells produce myelin sheath, which acts as an insulator andimproves the rate of information transmission along theaxon. The loss of these special cells significantly affectsaxonal propagation due to loss of myelin sheath, namely de-myelination [42]. Moreover, axonal regeneration is blockedby glial scar formation, which is a tissue barrier formed inthe injury site, as depicted in Fig. 1. In severe cases, axonalransmission is blocked due to demyelination and disruptionof the brain-spinal cord communication [42]. • Chronic injury:
This phase covers the days to years follow-ing the injury [40]. In this phase, injured axons experience aspecial kind of degeneration, called Wallerian degeneration,that disconnects the distal parts of the axons from their cellbodies while the glial scar continues to grow [43]. Since theinjury area is stable in the chronic phase, most treatmenttechniques are performed in this particular phase.The overall result of SCI mechanisms is the disruption ofaxonal pathways located within the spinal cord, which resultsin either the loss of sensory information transmission to thebrain or motor control signals to the muscles or both at thesame time [39]. Hence, re-establishing the brain-spinal cordcommunication is necessary to maintain the sensory and motorfunctions of the nervous nanonetwork. As discussed earlier,current treatment approaches fall in two main categories,namely the biological and the ICT-based treatment techniques.In the following, we discuss the major techniques that belongto these categories and the key differences between them.
C. Biological Treatment Techniques
Following the initial disruption of SCI, damage continuesto spread in the spinal cord in the secondary and the chronicphases. Biological treatment approaches aims to protect sur-viving neurons from the toxic post-injury environment, andto recover synaptic connections and healthy functionality bypromoting regeneration. On the contrary, ICT-based treatmenttechniques directly address the neural communication problemby either bypassing the injury site via external communicationinterfaces or by replacing the injured neurons by artificiallydesigned neurons.Next, we outline the existing fundamental biological treat-ment approaches and the challenges faced regarding theirimplementation. Furthermore, we discuss how these challengescan be addressed by ICT-based treatment techniques.Research efforts regarding biological treatments mainlyfocus on the following directions:
1) Neuroprotection:
Delivering drugs onto the injury siteis a possible method of neuroprotection. Recent research inthis area concentrates on drug delivery mediated by nanoma-terials (e.g., nanowires, nanoparticles, and carbon nanofibers),which have several advantages over conventional drug deliverymethods, such as providing targeted delivery and reachinginjury site by crossing the blood-spinal cord barrier [44]. Thechallenges in this direction include the following: • Application:
Devastating effects of secondary injury mech-anisms spread fast with the cascade of biological complexevents. Preventing this spread means intervening many bio-logical processes in the spinal cord with presumably numberof different neuroprotective agents. Thus, identifying theeffective drugs and their administrations in a limited timeperiod poses challenges. Moreover, there are many issuesregarding drug carrier design [44]. • Validation:
Several drugs are tested on animals and invitro environments for different injury scenarios. Identifyingeffective drugs and methods among many alternatives andverifying the efficacy by human trials still pose a challenge. • Functional outcome:
Although several neuroprotective drugsare reported to be resulting in some functional recovery [45],functional outcome is limited because complete recoveryalso necessitates the reformation of spinal cord circuitry.
2) Neural Regeneration:
Delivery of neuro-regenerativedrugs, cell-based therapy and tissue engineering aim to pro-mote neural regeneration. In drug delivery, growth factors,such as neurotrophins, help in improvement of nerve regener-ation, synaptic transmission, and plasticity [46]. Cell therapyefforts mostly focus on cell transplantation and grafting.Transplantation of stem cells is widely investigated as a cell-based therapy [47]–[49] since stem cells have differentiationcapabilities that allow them to change into neurons and glia.In addition, a variety of cells are used in studies, suchas Schwann cells, that are in-charge of producing myelinsheaths in peripheral nervous system and capable of improvingremyelination as well as axonal regrowth [50], [51]. Tissueengineering aims to promote regeneration of injured nerves bymeans of materials that can mimic natural scaffolding, suchas hydrogels, nanofiber scaffolds, and hybrid applications.Hydrogels are biocompatible networks of polymers that canimprove the environment for neural regrowth [43], [52], [53].Nanofiber scaffolds not only provide supportive scaffolding fordamaged cells but also deliver drugs and support transplantedcells [54]–[56]. Various combinations of hydrogels, nanofibers,and drugs are also studied for recovery of SCI as shown by[57]–[59]. Fundamental challenges in this direction include thefollowing: • Functional outcome:
Although the above-mentioned studiesreport some improvement regarding neural regeneration,and recent research show several initial results regardingfunctional neural network formation [60]–[63], directingregenerating neurons to form functional circuitry is stillchallenging [64]. • Human Trials:
Safety and efficiency of medicine treatmentand cell therapy on humans are still controversial [64],[65]. Regarding tissue engineering, functionally integratingbiocompatible materials to the tissues and the feasibility ofhuman trials pose potential challenges [66].
D. Biological versus ICT-based Treatments
Since the ICT-based treatment approaches do not haveprotective or supportive aims, they directly focus on com-munication problems between the brain and the spinal cord.Several initial studies have provided promising results for thesetechniques with regard to safety and the functional outcome.In this respect, some advantages over biological treatmentapproaches are stated below: • Functional outcome:
As processing, stimulation, and com-munication capabilities of the external interfaces advancewith the developments in ICT, restoration in the movements ecording of Neural Activity
From BrainFrom Tract
Processing Neural Signals
Neural Signals
Restoring Function
Assistive Neuroprosthesis
Stimulating Muscles
Stimulating Spinal Cord
Pattern Recognition Control Signals
Visual Feedback
Fig. 2: Basic structure of an NIS. of a subject with SCI can be observed as proved by a recentstudy [24]. • Human trials:
In addition to animal experiments, in studies,such as [25], promising improvements in terms of motorimpairment in human patients were observed. • Safety:
Although the long term safety of invasive recordingdevices in NIS is still not proven, there are less invasivealternatives with low risk neural interfaces, such elec-troencephalography (EEG)- [67] and electrocorticography(ECoG)-based systems.III. E
XISTING
ICT-
BASED T REATMENT T ECHNIQUES
In this section, we present prevalent ICT-based treatmenttechniques reported in the literature as well as future directionsthat may be possible in the NIS and AN frameworks. Twotechniques contrast with each other on the scale of theirapproaches with NIS focusing on macro-scale solutions andAN employing a micro/nano-scale approach.
A. Neural Interface Systems
While the aforementioned biological techniques aim to im-prove signal transmission problem through the injured spinalcord, NIS utilize nervous signals captured directly from thebrain or from the spinal cord before the injured part to restorethe motor function. Fig. 2 depicts the major components of atypical NIS, details of each are elaborated below.
1) Nervous recordings:
As mentioned, neural signals canbe either recorded from the brain [11], [19], [24], [27], [68]–[70] or the spinal cord [9]. While the former is more commonin literature, the latter can potentially provide signals withmore information from just a small anatomical region. Thekey factors in selecting a recording technique in an NISinclude efficacy, safety (particularly with regards to surgery),reliability, and longevity. Furthermore, ease of NIS training,support, cost, and availability are also important factors. In thefollowing, we review signal recording techniques from bothof the signal sources, and discuss their advantages as well aslimitations. a) Recording from the brain:
Brain signals can berecorded via non-invasive techniques, such as EEG, magne-toencephalography (MEG) or functional magnetic-resonanceimaging (fMRI). These techniques have low spatio-temporal resolution, low signal-to-noise ratio (SNR), and poor sensitiv-ity to high-frequency changes. Hence, they are insignificant forinvestigating the short-lived spatio-temporal dynamics of manybrain processes [71]. However, the feasibility of these tech-niques in detecting motor intent to some extent is demonstratedin literature, such as [19], [27], where robotic exoskeletonsare operated with use of control signals extracted by EEG andelectro-oculography recordings.In contrast, recordings with exceptionally high SNR, lowersensitivity to artifacts than EEG, and high spatio-temporalresolution can be achieved by invasive recording techniquessuch as ECoG and intracortically electrode placements [71].In ECoG, electrodes are implanted subdurally on the surfaceof the brain; hence, they are able to record the activityof groups of neurons and provide movement-related fieldpotentials. These recordings are robust over long periods [72],and higher spatio-temporal resolutions can be achieved byutilizing the micro-ECoG grids [73]. Furthermore, the ECoGsignal is used for offline decoding of seven degrees of freedomfor arm movements in monkeys and online decoding forindividual finger movements in human subjects [74], [75]. Thisdemonstrates that it can provide enough independent controlsignals for simple loco-motor task.However, the implantation of electrodes in the cortex isrequired to directly record neuronal action or local field po-tentials from the brain and to increase the spatial resolution ofthe recordings. This implantation yields to significant clinicalrisks as a result of infection or the formation of scar tissues.Hence, the electrodes need to be placed on a very small crosssectional area to minimize the damage to the tissue. Moreover,this recording technique requires sophisticated signal pro-cessing and computationally intensive algorithms to interpretthe neural activity and reliably separate signals of differentneurons [76]. Although point-and-click control is shown to besuccessfully achieved days after implantation of a -electrode array in an individual with tetraplegia [15], the long-term stability of the implanted electrodes need to be studiedin more depth. In particular, intracortical recordings faceshort-term recording failures due to implant micro-motion,mechanical mismatch of the device and tissue, foreign-bodyresponse, and formation of glial scar tissue that interfere withsignal transmission [77].
ABLE II: A list of methods used in a pattern recognition system.
Process Method Ref.
AutoRegressive component [79]Wavelet transform [80]Common spatial pattern [81], [82]
Feature extraction
Matched filtering [83]Principal component analysis [82], [84]
Dimension reduction
Independent component analysis [82], [84]Genetic algorithm [84]
Feature selection
Sequential selection [78]Linear discriminant analysis [82], [84]Support vector machine [82], [84]Bayesian statistical classifier [81], [85]K-nearest neighbor classifier [82], [85]
Classification
Artificial neural network [86] b) Recording from spinal tracts:
Although recordingelectrical signals from the brain is largely studied in theliterature, many technical challenges still exist in the use ofmicroelectrodes to reach a stable recording of individual cellactivities. The most important one is the formation of a layer ofactivated astrocytes around the recording electrode that makeslong-term recording of single spikes very difficult. Hence,recording motor control signals through cross-sectional areaof corticospinal tract is investigated in [9], [10]. The maindrawbacks of using this recording source for NIS can be listedas the following: • Designing a mechanically stable array of electrodes ischallenging. • The range of signal capture by an electrode array is limited,and thus the number of neural sources are far more than thearray recording electrode contacts. • The control signal cannot be received in corticospinal tractfor patients with tetraplegia or ALS.
2) Signal Processing:
Real-time processing of the recordedneural data needs to be done for deriving the intended taskby the patient. This process is seen as a pattern recognitionsystem that contains three main parts: (i) feature extraction; (ii)dimension reduction and feature selection; (iii) classificationor regression. First, the data is processed to derive a set offeatures. Feature sets extract the discriminating informationthat represent a dataset. Since the brain signals are the com-bination of several simultaneous sources and noise, extractionof an appropriate feature set is a challenging task. The datacan contain undesirable components, i.e., artifacts, which needto be removed to improve the performance of NIS. Moreover,not all of the recorded signals through multiple channels arerelevant for understanding the phenomena of interest. Hence,dimension reduction and feature selection methods are utilizedto remove irrelevant and redundant information. In Table II, alist of methods available for each step of a pattern recognitionsystem is provided [78].Apart from the aforementioned methods, a deep neuralnetwork (DNN), which can be trained by a training set of recordings, can also be utilized to extract the task intendedby a patient. The advantage of using a DNN is that itextracts the features by itself, which is a complex task asthe recorded neural data are high-dimensional and not well-known. Furthermore, the DNN predicts the non-linear systemsmore accurately [87].After extracting the intended task from neural recordings,control signals must be derived to restore or replace naturalfunction of a paralyzed or lost limb. In case of spinal cordstimulations for restoring the functionality of the injured spinalcord, these control signals are the parameters needed as theinput of the spinal cord stimulator. In [24], the extensor andflexor hotspots corresponding to different brain signals areidentified in intact monkeys, then the monkeys are paralyzed.Since such training data is not available in paralyzed patients,experiments on healthy subjects are needed to detect whetherthese hotspots are almost the same in different subjects. Evenif these hotspots are placed in different locations for eachsubject, the epidural stimulation method of [26] together witha machine learning (ML) algorithm can be used to gather therecorded data from healthy subjects and choose stimulationparameters [26].
3) Restoration of Function:
The extracted control signalsfrom the previous steps are then utilized to restore or replacenatural function of a paralyzed or lost limb. Such NIS includebrain-computer/machine/spine interfaces [7], [8], [12]–[25],[27], [78], or spinal cord-computer interfaces [9], [10] thatare described below: a) Operating assistive neuroprosthesis:
Real-time con-trol of robotic prosthesis and exoskeleton finds its applicationsin practical rehabilitation by replacing the lost motor functionto support of daily actions. Studies on using intra-corticalrecordings of the brain to control robotic prosthesis [16]–[18]demonstrate that paralyzed patients may recreate multidimen-sional control of complex devices even years after the injury.This is utilized in [19] to control a hand exoskeleton thathelps patients in restoration of independent daily life activities.Moreover, with sufficient trainings and use of EEG-controlledrobotic actuators, it is shown that patients achieve the abilityto perform voluntary motor control in key muscles belowthe spinal cord injury level [27]. However, using assistiveneuroprosthesis is not the focus of this study since it doesnot provide a solution for bypassing the injured part of thespinal cord to restore the control over muscles. b) Direct stimulation of muscles:
Direct stimulation ofmuscles by signals extracted from intra-cortical recordingswas first studied on monkeys with a transiently paralyzed arm[20]–[22], and then applied to a paralyzed human [25]. Inboth studies, the exact control signals for each of the mus-cles required in an activation are extracted and then appliedto stimulate the muscles. Results suggest that intra-corticalrecordings can be linked to muscle activation in real-time,enabling the control of muscles using the activity of neuronsin the motor cortex. This process essentially bypasses thespinal cord and restores the voluntary control of the paralyzedmuscles. However, electrical stimulation of muscles may cause
ENDRITES AXON
SOMA (a) Yellow light activates the opsin called halorodopsin turningthe neuron off.
SPIKES (b) Blue light activates the opsin called channelrhodopsin turningthe neuron on.
Fig. 3: Optogenetic, neuronal inhibition and excitation using light responsive proteins, called opsin [90]. fatigue, and long term stimulation by electrodes can damagemuscles. c) Spinal cord stimulation:
Epidural spinal cord stimula-tion and intraspinal microstimulation are two existing methodsin the literature for inserting motor control signals back to themotor circuitries. In the first one, the stimulating electrodes areplaced in the epidural space of the spinal cord [26], while theelectrodes are implanted within the ventral gray matter [88],[89] in the second method.The sensory feedback from the body, which is processed byneurons in the spinal cord, is crucial in controlling the motorcommands [91]. Hence, the idea of stimulating the spinalcord just enough to make it sensitive and responsive to thesensory input is studied in [26]. An array of electrodes isimplanted in epidural space of the spinal cord of an individualwith a clinically complete motor injury. These electrodes areplaced over the identified extensor and flexor hotspots, and apulse generator is used to control the stimulation parameters.Different set of stimulating parameters are then sent to thepulse generator to observe the response of the patients bodyto the stimulation. The epidural stimulation in [26] resultedin locomotor-like patterns and full weight-bearing standingwith assistance provided only for balance. Instead of selectionof stimulating parameters by a physician, which is done in[26], the recorded signals from brain or spinal cord can beprocessed by a processing unit to set these parameters. Usingbrain signals to control the epidural stimulation is studiedin [24], [92], where locomotor movements, namely walkingand climbing, are achieved in paralyzed monkey and rat,respectively.In [23], intra-spinal micro-stimulation is used to controlhand movement in monkeys whose hands were temporarily paralyzed. This stimulation method provides more naturalrecruitment order of motor units and reduces the numberof required electrodes and controllers compared to epiduralstimulation. However, higher risk of tissue damage exists dueto implantation of electrode into the spinal tissue. Implantedelectrodes retain a scar tissue, and long-term stability andsafety of the implanted electrodes are some of the majorchallenges [93].By using current electrical stimulation techniques in eitherepidural stimulation or intra-spinal micro-stimulation, cell-specific activation of spinal cord’s motor circuitry is challeng-ing due to interference from non-target neuron populations.In this respect, optogenetics can be a key to control thelimbs by activating or suppressing the specific population ofneurons in the spinal cord as shown in Fig. 3 [94], [95].Open issues include finding encoding methods for motorcommands recorded from the brain in order to deliver mean-ingful signals to the spinal cord and developing proper lightdelivery techniques for spinal cord [96]. The former requiresidentifying contributions of different neuron populations on themotor control. Regarding the latter, current research focuses onthe RF-powered wireless systems [97], [98] since employingoptical fibers in the spinal cord is not possible. This is contraryto the case of the brain since penetrating an optical fiber in thespinal cord requires the severing of white matter tracts, whichcarry high information density and have minimal redundancy;thus, local damage can have global consequences [96].Anotherapproach could be designing an ultrasound-powered optoge-netic stimulator system, such as those shown in [99], [100],which can reach deeper layers in the spinal cord, to employthem in the spinal cord. re-synaptic RecordingTransducer(Passive electrodesor BioFET) ProcessingUnit StimulatingTransducer After-post-synapticArtificial NeuronNeuron Neuron Fig. 4: General architecture of an AN.
B. Artificial Neurons
Another potential direction for the treatment of SCI couldbe the use of artificially-created neurons to replace the injuredbiological neurons. If perfectly implemented, such an approachcan potentially reverse the loss of any function occurred dueto SCI; however, the associated challenges are not negligibleby any means either. The general architecture of an artificialneuron (AN) is shown in Fig. 4, where a presynaptic neuronconnects to a processing unit by means of a transducer.The AN performs processing actions similar to its biologicalcounterparts, and thus completes the nervous pathways.Some key characteristics required by a functioning AN thatmay replace biological neurons are outlined below: • Plasticity:
The AN should be able to change synapticweights in a plastic manner since plasticity forms the basisof memory and cognitive functions of the nervous system. • Implementation scale:
The scale of implementation decidesthe level on which the system behaves similar to a biolog-ical neuron. From the perspective of a single neuron, thescale can either be synaptic or neuronal. On the synapticscale, individual synapses can be identified and interfacedto synapses from other neurons, whereas on the neuronalscale, the neuron as a whole may be interfaced. Bothsystems require a different set of interfacing techniques butthe complexity and control available in AN systems willincrease against a decrease in the scale. • Biocompatibility:
Same as NIS, the AN needs to providelifelong stability, which requires biocompatibility and cyto-compatibility of the device, i.e., the device must integrateproperly within the body and neural tissues, and provide theprolonged maintenance of the desired performance [101]. • Physical dimensions:
The size of ANs need to be definedbased on the biological neurons that they will be integratedwith. Thus, their dimensions must be on the scale of nmwhen aimed to be used in synaptic clefts while they needto have larger dimensions when used together with spinalneurons. • Energy requirements:
Energy self-reliance is another majorissue for any implanted system. Use of batteries may notbe a viable solution for long-term operation. The energyrequired by an AN can be harvested from the body heator blood pressure by creating gradients within the body[102]; however, other novel methods can be proposed. For instance, ANs may be designed such that they can harvestenergy directly from the energy sources of the body, suchas glucose, and their efficiency should also be considered. • Amplification:
Processing of biological inputs may requirethe use of amplifications since the transmission needs to becarried over some distance and the typical energy of signalsis on the order of µ J. Such amplifiers must have low noise,low power, and high precision characteristics. • Implanting process:
This process is almost as critical asthe development of ANs. Considering the scale of neurons,operation by a human surgeon is not a viable option.The implant technologies may require the use of roboticsurgeries that are capable of operating at n m scales. Insuch cases, the system should be intelligent since the sheernumber of decisions will be too large for a human to process.The system would need to automatically identify differentsections of biological neuron by means of imaging, andproceed with the surgery by connecting the appropriate ANsections accordingly. Thus, apart from the robotic arms andsurgery tools, the implantation requires significant ML andartificial intelligence (AI) systems.The implantation problems elaborated above can be miti-gated by using systems that deploy themselves by means ofself-organization.The characteristics/requirements of ANs pose significantchallenges; however, once developed, the benefits of suchsystems are also unlimited. Apart from the treatment of SCI,they may be utilized to develop computing systems similar tothe human brain. Additionally, they may be used as bio-cyberinterfaces for applications targeting the nervous system.To date, there is no work that may be considered as areplacement for a biological neuron. Thus, we discuss somemajor studies that are approaching to the problem in a similarmanner. Two major directions reported, namely neuromorphicand biomimetic neurons, are explained below:
1) Neuromorphic neurons:
Neuromorphic neurons arebased on electronic circuits, which particularly perform neu-romorphic computation, where very-large-scale integration(VLSI) of individual neurons or synapses are utilized in asimilar manner to the human brain [103].The biggest advantage of neuromorphic systems is that theyinclude plasticity as the basis of their models; however, im-plementation of ANs by neuromorphic methods faces severalottlenecks, such as huge size, high-power requirements, andinterfacing issues. Even for some of the newer devices thatuse low-power and have small sizes, interfacing issues remainsince neuromorphic systems are not designed with neuralinterfaces [104].
2) Biomimetic neurons:
Biomimetic circuits mimic bio-chemical processes using synthetic materials [105]. Since mostcommunication between biological entities occurs by meansof molecule exchange and biochemical reactions of thesemolecules, biomimetic circuits are ideal candidates for directinterfacing with biological systems. They are employed to re-alize applications, such as artificial muscles, smart membranes,biological transducers, and brain-machine interfaces [105].The basic function of a neuron is realized by meansof enzyme-based amperometric biosensors and organic ionpumps in a biomimetic fashion by [106]. Although this work isbio-compatible and the detection of bio-molecules is achievedon the scales similar to those of biological neurons, manyof the characteristics expected from an AN are not met. Forinstance, there is no plasticity in the system, macro-scaledimensions are used [106], and the system is tested in vitrowithout any interface to the biological neurons.Similar studies that include plasticity and direct interfacewith biological neurons may be the first step towards thedevelopment of organic biomimetic ANs. Size reduction and invivo testing may follow afterwards. Another direction could bethe use of plasticity designs from neuromorphic systems basedon biomimetic implementations. This may be the approach thatis most successful and has the least time-to-market.IV. P
ROPOSED A PPROACHES OF THE
ICT-
BASED T REATMENT OF
SCIIn this section, we introduce two potential approaches forthe treatment of SCI by considering the current literature andthe desired characteristics of NIS and ANs.
A. Neural Interface Systems with Enhanced Feedback
The main technical challenges in using an NIS to enablethe ICT-based treatment of SCI are over-viewed below: • Providing somatosensory feedback : It is shown that visualmonitoring of a limb’s motion can partially correct the short-comings in movement of patients with large-fiber sensoryneuropathy. These subjects face slow and uncoordinatedmovement as a result of the somatosensory feedback for nor-mal motor control [107]. Hence, patient’s vision provides afeedback to the NIS. However, the effect of visual feedbackon movements is not as fast as proprioceptive system, whichthen causes instability in movement [108]. Thus, utilizingproprioceptive feedback or a feedback from spatio-temporalcharacteristics of limb’s in an NIS can assist patients toperform more natural movements and also feel stimulatedlimbs or neuroprosthesis [92], [109]. As an example, stim-ulating peripheral afferent nerves in the limb’s stump withintrafascicular electrodes [110], [111] or cuff-like electrodes[112] is studied in the literature for providing the naturalsensation of missing limbs in amputee subjects. However, these methods are not applicable in paralyzed patients, whohave also lost the sensory pathway. In such cases, the tactileand proprioceptive feedback must be provided by artificiallystimulating the sensory cortex. • Appropriate neural recording technique, spinal cord, andbrain stimulation : Lifelong stable neural signal recordingsand stimulation are required for which the important fac-tors are reliable chronic recordings, biocompatibility andcytocompatibility of the device, the ability of the device tointegrate properly within neural tissues, and the prolongedmaintenance of desired electrical properties. Furthermore,both recording and stimulating devices must have highspatio-temporal resolution to handle the required degree offreedom for restoring function after the SCI. Graphene isa bio-compatible 2D material, and its sensitivity to dif-ferent ions can be modified by various etching methodsand functionalization [113]. Hence, it can be used forsensing fluctuations in specific ion concentrations. Grapheneis electrically conductive, and it can be patterned intonanoscale, electrically disconnected, conductive patches viahydrogenation, etching or fluorination [114], [115]. Thismakes it suitable for implementing high resolution neuralinterfaces which can both generate and differentiate thehighly localized fluctuations in ion concentrations. • Fully implantable system : To offer more comfort for patientsand reduce the risk of inflammation, the NIS must be fullyimplantable. Moreover, the system must provide the requiredbandwidth for transmitting data between its different com-ponents while considering the power consumption.To address the above-mentioned challenges, we propose acomprehensive framework for realizing an NIS with enhancedfeedback,
EF-NIS , that utilizes the neural signals recordedfrom the brain and stimulates the spinal cord for restoringmotor functions in the paralyzed patients as shown in Fig. 5.Components of the proposed EF-NIS are explained below:
1) Recording device and the communication interface:
Thegraphene-based high resolution neural interface (GNI) willbe placed over the primary motor cortex to record the brainactivities of the patient. This recording device is implantedtogether with a processing unit and a transmitter, which reportsthe recorded signals in a wireless manner.The bandwidth demand of the wireless transmission channelbetween the brain implant and the external processing unit iscorrelated with the number of recording channels employed.For better estimation of brain commands, we need denselydistributed recording sites that require high data rates (onthe order of Mbps) [116]. On the other hand, the transmittershould dissipate low energy because even 1 ◦ C of temperatureincrease can damage neural tissues [117]. Considering theselimitations, ultra-wide band (UWB: . - . GHz) systems,which can provide high data rate within allowable powerdissipation range with the minimal chip area [118], is themost suitable solution [119]. Therefore, the communicationinterface between the brain implant-processing unit of the EF-NIS consists of a simple UWB transmitter at the brain implant timulatorSinkFeedback Path DNNProcessingUnitWSN TxRxRxGraphene-based high-resolution Tx UWBGNI Tx Rx DNN Tx Rx SitumulatorUWBPulseGenerator ControlSignals UWBUWB Forward PathFeedback PathNeural Interface (GNI)WSN Sink Tx Rx DNN Tx Rx Brain SitumulatorUWB UWBDNN ProcessingUnit RxTx UWBControlSignals UWB TxStimulatorBrain Forward PathRx
Fig. 5: EF-NIS, a neural interface systems with enhanced feedback. and a more complex UWB receiver at the processing unit[120]. Furthermore, compression techniques with efficient im-plementation, similar to one in [121], can be utilized to reducedata rate without considerably increasing the complexity.
2) Processing unit and the spinal stimulator interface:
Theprocessing unit performs two tasks: (i) decoding the intent ofthe subject from neural signals; (ii) generating control signalsfor spinal cord stimulation. This unit will be placed in arechargeable and mobile device. Hence, it needs a wirelesstransmitter to transfer the stimulation parameters to the siteof spinal cord stimulation. The transmission between theprocessing unit and the spinal cord neural stimulator canbe performed at lower data rates compared to the brainimplant-to-processing unit link since the dimensionality of thetransmitted signal reduces. Additionally, the receiver at theneural stimulator must operate with low energy [124]. Area-and power-efficient UWB transmission is suitable for this link[125], [126]. Moreover, it is also advantageous in terms ofinterference [127].The received control signals at the spinal cord stimulatorare then used for setting the stimulation parameters. TheGNI is able to both recognize and generate fluctuations inconcentration of specific molecules in a local area. Hence, itwill also be used for high resolution stimulating of nerve cellslocated in the spinal cord.
3) Muscle to brain communication interface (Feedback):
Inliterature, there is a variety of wearable sensors for monitoringthe body activity. Among those, especially flexible tattoo-based and skin-like sensors are profitable due to their soft andultra-thin designs that provide comfort for users. Furthermore,they are generally made of bicompatible materials, such as (a) Graphen electronic tattoo sensors [122].(b) A transparent graphene-based piezoelectric motion sensor [123].(c) Flexible motion sensor using gold thin films [123].
Fig. 6: Examples of flexible tattoo-based and skin-like sensors. raphene [122], [123], [128], [129] and gold [130], as shownin Fig. 6, which makes them suitable for long-term use.These sensors are also capable of recording multiple sensorymodalities, such as muscle activity and touch [122]. For thecommunication interface, wired, wireless or hybrid approachescan be followed. For high density of sensors, which is thecase of multiple modalities being recorded, we can use wiredsensor arrays, whose recordings are gathered at the sink node,which has a direct wireless communication interface withthe processing unit or with the brain implant. Otherwise,communication modules of the sensors can be integrated toNear Field Communication (NFC) and Radio Frequency Iden-tification (RFID) technologies, which are enabled by flexiblethin antennas, such as graphene-based antenna shown in [131].An alternative to wearable sensors is ultrasound-based im-plantable neural dust, which is validated in rat peripheralnerves and is shown to be capable of recording neural activityin the peripheral nervous system [132]. This system usesbackscattering and other advantageous aspects of ultrasound,such as low attenuation through the tissues and low energyconsumption compared to RF technologies currently employedin implantable devices [133]. In addition, further miniaturiza-tion is needed in this system similar to the aforementionedflexible sensor-based systems.The collected sensory information is then transmitted to theprocessing unit to generate the parameters for somatosensorycortex stimulation. For this aim, the responsible regions ofsomatosensory cortex for sensory feedback from each legshould be identified. This task can be achieved by collectingtactile and muscle activities in healthy subjects, and recordingthe corresponding activities in somatosensory region duringwalking. Then, a DNN is trained to estimate the activity ofthe somatosensory cortex according to the collected sensoryinformation. Since we consider the SCI injuries, in whichthe sensory pathway is also damaged, EF-NIS utilizes thisinformation to generate stimulation parameters that activatesimilar somatosensory regions in patients.In the next step, the processing unit transmits the stimulationparameters to the controller unit that is surgically placedover the pia mater, i.e., the delicate innermost layer of themeninges that is the membranes surrounding the brain andthe spinal cord. The transmission is again performed throughthe UWB communication interface. The controller sends thecontrol signals to the graphene-based high-resolution neuralinterface, located over the somatosensory region, to stimulatethe brain through manipulating ionic concentration levels.
B. Self-Organizing Artificial Neurons
Although the preceding neural interface is relatively easyto implement and may return a high degree of motion tothe patients, the movements will not be precise. In similar,although the nervous system’s information processing capacityincreases by the use of an NIS, its communication rate is not ashigh as before the injury. Before an SCI, being normally wired,the nervous system can be assumed to operate at optimumconnectivity with maximum communication rate. However, after the SCI, while using an NIS, neural information canonly be communicated at limited neural regions, a setup farfrom the optimum connectivity, which results in diminishedcommunication rate. This materializes itself, for instance, aslimited stimulation of some estimated areas of the spinal cordfor the motor movement. Apart from limitations in establishingconnectivity, operating through interfaces located far away,e.g., the cerebral cortex, from the spinal injury site, where thelost connectivity to be reestablished resides, additionally suf-fers from the delocalization of neural signals as they propagate.Thus, establishing a simple lost connection at the injury sitemay correspond to collecting readings from the multiple sitesof cortex together with some signal post-processing. However,the delocalization process, coupled with a lack of completeneural signals, results in unrecoverable information loss, andthe lost connection will not be recovered in full effect. On theother hand, delocalization of neural signals is not an issue ifone opts to work at the injury site.A solution to this problem could be the retrieval of connec-tions lost at the injury site via the self-organizing ANs. Theconnectivity, and consequently communication rates, could be(at least theoretically) brought back to the pre-injury levels byemploying artificial nervous connections across the injury site,where the signals are already localized in specific positions.An architecture, where ANs with neuromorphic properties aresuspended in a hydrogel, could be employed for such purpose.The hydrogel-suspended network of ANs can form bridgesacross the injury site and make new nervous connectionswith biological and artificial neurons via self-organization. Arepresentative diagram is shown in Fig. 7. Below, we discussthe major properties of the proposed self-organizing AN:
1) Plasticity:
In order for an AN to adapt to its envi-ronment, plasticity should be introduced in the system andrealized by utilizing neuromorphic designs. The firing rate ofeach AN depends on the signals it receives from synapseswith other neurons that may be both biological or artificial.Each synapse of the AN is associated to a variable
W eight that models plasticity of the synapse and retains the weightof a particular synapse [134]. The
W eight may function tostrengthen or weaken depending on the synaptic connectiontype (excitatory or inhibitory). The overall effect of plasticityis the introduction of memory in the system, which is a keycharacteristic of the nervous system.
2) Self-organization:
Since functional ANs are required tointerface with biological neurons, their dimensions are dictatedby the task. If ANs are communicating on axonal and somaticlevels, their sizes will be micro-scale whereas if ANs aim tocommunicate on synaptic scales, their sizes should be nano-scale. As discussed earlier, the implantation of AN systemspropose a major challenge since surgery on the micro/nano-scale itself is an open problem. This can be bypassed byconsidering systems that use self-organization to create the ANbridge across the nervous injury. Such a system is shown inFig. 7, where AN bridges form across the injury site over time.Self-organization is a hot research area, where local in-teractions between individual entities give rise to the overall a) Before self-organization. (b) After self-organization.
Fig. 7: Diagram of the self-organizing ANs suspended in a hydrogel. structure of the system. The simplest approach towards thisgoal is the use of electric or magnetic fields. Since the actionpotentials of the nervous system are themselves electrical sig-nals, self-organization can be based on its use as the attractorfield. Graphene nanosolenoids discussed in [135] may providethe basis for an implementation in this regard, both in terms ofthe required size and low energy requirements and magneticproperties. The nanosolenoids, however, are not sufficient bythemselves for self-organization, because the nervous signalsare not continuous in nature and the AN bridge may fallapart if a signal is absent for some time. To ensure long-term endurance of the structures, we either need mechanicalstructures, such as hinges [136], or the use of magnetic fieldsexternal to the body [137].
3) Energy Harvesting:
For the system to be completelyindependent, ANs need to collect energy from their vicinity.This energy can be scavenged within the body [102] or throughexternal sources using wireless power transfer (WPT).Energy harvesting (EH) within the body can either be basedon biochemical or biomechanical processes [138]. Chemicalreactions within the body, such as glucose uptake and fooddigestion, can be considered as the sources for biochemical EHmechanisms. On the other hand, biomechanical EH can eitherbe based on voluntary body functions, such as movement, orinvoluntary functions, such as blood flow, breathing, and heartbeat. The key advantage of EH within body is the fact thatthe designed system can be independent of external influences,allowing subjects to move more freely.WPT can be realized by providing power to the systemexternally by means or either inductive or magnetic coupling,or by means of electromagnetic radiation. Apart from reducingthe challenges of designing EH mechanisms in an alreadycomplex scenario of SCI treatment, if an external source isselected, the same may be used for the self-organization of the AN devices, and as a readout interface of the AN network.Finally, it should be noted that EH is a very active areaof research. New efforts are focused on reducing the powerrequirements of networks by using energy-efficient protocols,conservation schemes, and novel topologies. Future work canfurther elaborate on the directions overviewed here.V. F
URTHER APPLICATIONS OF THE
ICT-
BASEDTREATMENT OF
SCIBy revisiting the current treatment techniques of SCI froman ICT perspective, we can provide rigorous optimizationframeworks for the improvement of these techniques. How-ever, apart from SCI treatment, such ICT-based techniques canhave applications in a number of other scenarios. Thus, in thissection, we discuss some key applications that are realizableas byproducts of ICT-based SCI treatment techniques.
A. The Internet of Bio-Nano Things
The Internet of Bio-Nano Things (IoBNT) is an emergingsubfield of IoT, which focuses on developing artificial biolog-ical/bioinspired nano-scale devices capable of communicatingwith each other and their surroundings to establish predefinedtasks at the nano-scale, with envisioned applications rangingfrom intra-body sensing and actuation networks to environ-mental control of chemical and biological agents and pollution[139]. In the case of intra-body applications, experience gainedfrom techniques developed for SCI treatments, in particular,molecular NIS, can provide the means for establishing com-munication with nano-networks deployed into the body. Suchapplications include remote monitoring of the body for variousvectors and automatized targeted drug delivery upon detectionof anomalies in observed data. . Neurophysiology/Physiology
Large amounts of data can be retrieved from the nervoussystem by using the interfaces of the proposed SCI treatments.These data can be processed to obtain a mapping betweenthe architecture of the underlying network and its function,and contribute vastly to our understanding of biological neuralnetworks. On the other hand, NIS-like systems equipped withmolecular sensors capable of sensing specific biomoleculescan be used to interface other biological systems, such ascardiovascular and lymphatic systems. Data collected fromsuch an endeavor will significantly contribute to our under-standing of human physiology. For instance, we can studynervous-related diseases, such as Alzheimer’s disease, multiplesclerosis, and Parkinson’s disease, in unprecedented detailusing electrical/molecular NIS as interfaces to the centralnervous system. This may lead to novel disease diagnosis andtreatment techniques as well as the identification of key factorsthat cause such conditions.
C. Neuromorphic Computing
A firm understanding of the relation between the architec-ture and function of a neural network would allow us to designour own artificial neural networks (ANNs, as a subtopic ofML and AI) with specialized functions, and employ thesearchitectures for neuromorphic computation. Neuromorphicarchitectures process data in a highly parallel manner asopposed to the serial structure used by von Neumann archi-tectures employed by contemporary computers. This allowsANNs to process data at comparatively very low frequencies,and therefore to be highly energy efficient.One application of functional ANNs lies in their integrationinto neural interfaces for on-body preprocessing of collecteddata, which is necessary as the volume of raw data collectedby high spatio-temporal resolution interfaces will be so largethat transmission of this data without any preprocessing isdoomed to have heating problems, which is not desired for anon-body device. Thus, we will have a positive feedback loop:Understanding gained from neural interfaces will advance ourcapability of neuromorphic computing, which, in return, willpave the way for more advanced interfaces.
D. Augmented Humans (Cyborgs)
The use of external machinery to enhance capabilities of ahuman body is not new. In fact, prosthetics are widely usedto help people with missing limbs to regain some function-alities lost due to suffered injury. With advancements in SCItreatment technology, it is expected that such prostheses willbe upgraded from being mere mechanical tools to interactiveelectronic devices capable of communicating with humannervous system as the so-called neural prostheses. Interest-ing applications in such directions may include extrasensoryfeedbacks to achieve new sensations, mechanical body suitsthat allow superhuman strength, direct integration of gamingin the human body, among others. VI. C
ONCLUSION
SCI are serious health conditions that adversely affect thelife of a plethora of patients around the world. Existing biolog-ical methods not only aim to re-establish axonal connectionby promoting nerve regeneration but also minimize harmfultoxicities of the injury mechanisms. However, to date, noexisting treatment technique has shown significant success.Among the ICT-based systems, NIS simply bypass the injurysite by decoding the brain signals and providing them tothe recipient muscles. On the other hand, employment ofANs aims to compensate the functional loss by replacing theinjured neurons with engineered neurons. Considering these,we propose two novel approaches, namely EF-NIS (NIS withenhanced feedback) and self-organizing ANs, and identify keyareas that may be most important in this regard. For NIS, apartfrom the need of high resolution recording and stimulationsetups, we identify that the feedback from impaired limbs backto the brain is also very essential. For ANs, we propose the useof self-organization to tackle the transplant issues in a smarterway and also EH to make the devices fully independent. Thedevelopment of these treatments will have a great impact ona number of application areas, such as IoBNT.R
EFERENCES[1] O. B. Akan et al. , “Fundamentals of molecular information andcommunication science,”
Proceedings of the IEEE , 2016.[2] S. G. Waxman,
Clinical Neuroanatomy . New York: McGraw-Hill,2013.[3] S. L. James et al. , “Global, regional, and national burden of traumaticbrain injury and spinal cord injury, 1990–2016: a systematic analysisfor the global burden of disease study 2016,”
The Lancet Neurology ,vol. 18, no. 1, pp. 56–87, 2019.[4] “Spinal cord injury (sci) 2016 facts and figures at a glance,”
TheJournal of Spinal Cord Medicine , vol. 39, no. 4, pp. 493–494,2016, pMID: 27471859. [Online]. Available: http://dx.doi.org/10.1080/10790268.2016.1210925[5] E. E. Fetz, “Real-time control of a robotic arm by neuronal ensembles,”
Nature neuroscience , vol. 2, pp. 583–584, 1999.[6] J. Wessberg et al. , “Real-time prediction of hand trajectory by ensem-bles of cortical neurons in primates,”
Nature , vol. 408, no. 6810, pp.361–365, 2000.[7] M. D. Serruya et al. , “Brain-machine interface: Instant neural controlof a movement signal,”
Nature , vol. 416, no. 6877, pp. 141–142, 2002.[8] D. M. Taylor et al. , “Direct cortical control of 3d neuroprostheticdevices,”
Science , vol. 296, no. 5574, pp. 1829–1832, 2002.[9] Y. Guo et al. , “Encoding of forelimb forces by corticospinal tractactivity in the rat,”
Frontiers in neuroscience , vol. 8, 2014.[10] S. Gok and M. Sahin, “Prediction of forelimb muscle emgs from thecorticospinal signals in rats,” in
Engineering in Medicine and BiologySociety (EMBC), 2016 IEEE 38th Annual International Conference ofthe . IEEE, 2016, pp. 2780–2783.[11] P. R. Kennedy and R. A. Bakay, “Restoration of neural output froma paralyzed patient by a direct brain connection,”
Neuroreport , vol. 9,no. 8, pp. 1707–1711, 1998.[12] P. R. Kennedy et al. , “Direct control of a computer from the humancentral nervous system,”
IEEE Transactions on rehabilitation engineer-ing , vol. 8, no. 2, pp. 198–202, 2000.[13] S. Musallam et al. , “Cognitive control signals for neural prosthetics,”
Science , vol. 305, no. 5681, pp. 258–262, 2004.[14] L. R. Hochberg et al. , “Neuronal ensemble control of prosthetic devicesby a human with tetraplegia,”
Nature , vol. 442, no. 7099, pp. 164–171,2006.[15] J. Simeral et al. , “Neural control of cursor trajectory and click bya human with tetraplegia 1000 days after implant of an intracorticalmicroelectrode array,”
Journal of neural engineering , vol. 8, no. 2, p.025027, 2011.16] L. R. Hochberg et al. , “Reach and grasp by people with tetraplegiausing a neurally controlled robotic arm,”
Nature , vol. 485, no. 7398,pp. 372–375, 2012.[17] J. L. Collinger et al. , “High-performance neuroprosthetic control byan individual with tetraplegia,”
The Lancet , vol. 381, no. 9866, pp.557–564, 2013.[18] M. Velliste et al. , “Cortical control of a prosthetic arm for self-feeding,”
Nature , vol. 453, no. 7198, pp. 1098–1101, 2008.[19] S. Soekadar et al. , “Hybrid eeg/eog-based brain/neural hand exoskele-ton restores fully independent daily living activities after quadriplegia,”
Science Robotics , vol. 1, no. 1, p. eaag3296, 2016.[20] C. T. Moritz et al. , “Direct control of paralysed muscles by corticalneurons,”
Nature , vol. 456, no. 7222, pp. 639–642, 2008.[21] E. A. Pohlmeyer et al. , “Toward the restoration of hand use to aparalyzed monkey: brain-controlled functional electrical stimulation offorearm muscles,”
PloS one , vol. 4, no. 6, p. e5924, 2009.[22] C. Ethier et al. , “Restoration of grasp following paralysis through brain-controlled stimulation of muscles,”
Nature , vol. 485, no. 7398, pp.368–371, 2012.[23] J. B. Zimmermann and A. Jackson, “Closed-loop control of spinalcord stimulation to restore hand function after paralysis,”
Frontiers inneuroscience , vol. 8, p. 87, 2014.[24] M. Capogrosso et al. , “A brain–spine interface alleviating gait deficitsafter spinal cord injury in primates,”
Nature , vol. 539, no. 7628, pp.284–288, 2016.[25] C. E. Bouton et al. , “Restoring cortical control of functional movementin a human with quadriplegia,”
Nature , vol. 533, no. 7602, pp. 247–250, 2016.[26] S. Harkema et al. , “Effect of epidural stimulation of the lumbosacralspinal cord on voluntary movement, standing, and assisted steppingafter motor complete paraplegia: a case study,”
The Lancet , vol. 377,no. 9781, pp. 1938–1947, 2011.[27] A. R. Donati et al. , “Long-term training with a brain-machine interface-based gait protocol induces partial neurological recovery in paraplegicpatients,”
Scientific reports , vol. 6, 2016.[28] E. Balevi and O. B. Akan, “A physical channel model for nanoscaleneuro-spike communications,”
IEEE Tran. Com. , vol. 61, no. 3, pp.1178–1187, 2013.[29] D. Malak and O. Akan, “Communication theoretical understanding ofintra-body nervous nanonetworks,”
IEEE Communications Magazine ,vol. 52, no. 4, pp. 129–135, 2014.[30] ——, “A communication theoretical analysis of synaptic multiple-access channel in hippocampal-cortical neurons,”
IEEE Tran. Com. ,vol. 61, no. 6, pp. 2457–2467, 2013.[31] H. Ramezani and O. B. Akan, “A communication theoretical model-ing of axonal propagation in hippocampal pyramidal neurons,”
IEEETransactions on NanoBioscience , 2017.[32] ——, “Synaptic channel model including effects of spike width varia-tion,” in . ACM, 2015.[33] ——, “Importance of vesicle release stochasticity in neuro-spike com-munication,” in
IEEE, 2017.[34] H. Ramezani et al. , “Rate region analysis of multi-terminal neuronalnanoscale communication channel,” in
IEEE,2017.[35] H. Ramezani and O. B. Akan, “Information capacity of vesicle releasein neuro-spike communication,”
IEEE Communications Letters , 2017.[36] T. Khan et al. , “Diffusion-based model for synaptic molecular com-munication channel,”
IEEE Transactions on NanoBioscience , 2017.[37] N. A. Abbasi and O. B. Akan, “A queueing-theoretical delay analysisfor intra-body nervous nanonetwork,”
Nano Com. Net. , 2015.[38] N. A. Abbasi et al. , “Controlled information transfer through an invivo nervous system,”
Scientific reports , vol. 8, no. 1, p. 2298, 2018.[39] J. W. McDonald and C. Sadowsky, “Spinal-cord injury,”
The Lancet ,vol. 359, p. 417425, Feb 2002.[40] S. N. A. et al. , “From basics to clinical: A comprehensive review onspinal cord injury,”
Progress in Neurobiology , vol. 114, p. 2557, 2014.[41] R. Papastefanaki, FlorentiaMatsas, “From demyelination to remyelina-tion : The road toward therapies for spinal cord injury,”
Glia , vol. 63,no. 7, pp. 1101–1125, 2015.[42] C. A. Oyinbo, “Secondary injury mechanisms in traumatic spinal cordinjury: a nugget of this multiply cascade.”
Spinal Cord , vol. 71, pp.281–299, 2011. [43] S. Kabu et al. , “Drug delivery, cell-based therapies, and tissue engineer-ing approaches for spinal cord injury,”
Journal of Controlled Release ,vol. 219, pp. 141–154, 2015.[44] J. Y. Tyler et al. , “Nanomedicine for treating spinal cord injury,”
Nanoscale , vol. 5, no. 19, pp. 8821–8836, 2013.[45] A. Faccendini et al. , “Nanofiber scaffolds as drug delivery systems tobridge spinal cord injury,”
Pharmaceuticals , vol. 10, no. 3, p. 63, 2017.[46] S. M. Willerth and S. E. Sakiyama-Elbert, “Approaches to neural tissueengineering using scaffolds for drug delivery,”
Advanced drug deliveryreviews , vol. 59, no. 4, pp. 325–338, 2007.[47] P. Lu et al. , “Long-distance growth and connectivity of neural stemcells after severe spinal cord injury,”
Cell , vol. 150, no. 6, pp. 1264–1273, 2012.[48] H. Nakajima et al. , “Transplantation of mesenchymal stem cells pro-motes an alternative pathway of macrophage activation and functionalrecovery after spinal cord injury,”
Journal of neurotrauma , vol. 29,no. 8, pp. 1614–1625, 2012.[49] J. Sharp et al. , “Human embryonic stem cell-derived oligodendrocyteprogenitor cell transplants improve recovery after cervical spinal cordinjury,”
Stem cells , vol. 28, no. 1, pp. 152–163, 2010.[50] H. Kanno et al. , “Combination of engineered schwann cell grafts tosecrete neurotrophin and chondroitinase promotes axonal regenerationand locomotion after spinal cord injury,”
Journal of Neuroscience ,vol. 34, no. 5, pp. 1838–1855, 2014.[51] D. D. Pearse et al. , “Transplantation of schwann cells and/or olfactoryensheathing glia into the contused spinal cord: Survival, migration,axon association, and functional recovery,”
Glia , vol. 55, no. 9, pp.976–1000, 2007.[52] P. Krsko et al. , “Length-scale mediated adhesion and directed growthof neural cells by surface-patterned poly (ethylene glycol) hydrogels,”
Biomaterials , vol. 30, no. 5, pp. 721–729, 2009.[53] M. J. Mahoney and K. S. Anseth, “Three-dimensional growth andfunction of neural tissue in degradable polyethylene glycol hydrogels,”
Biomaterials , vol. 27, no. 10, pp. 2265–2274, 2006.[54] Y. Liu et al. , “A self-assembling peptide reduces glial scarring, atten-uates post-traumatic inflammation and promotes neurological recoveryfollowing spinal cord injury,”
Acta biomaterialia , vol. 9, no. 9, pp.8075–8088, 2013.[55] J. Guo et al. , “Reknitting the injured spinal cord by self-assemblingpeptide nanofiber scaffold,”
Nanomedicine: Nanotechnology, Biologyand Medicine , vol. 3, no. 4, pp. 311–321, 2007.[56] V. M. Tysseling-Mattiace et al. , “Self-assembling nanofibers inhibitglial scar formation and promote axon elongation after spinal cordinjury,”
Journal of Neuroscience , vol. 28, no. 14, pp. 3814–3823, 2008.[57] B. Chen et al. , “Repair of spinal cord injury by implantation of bfgf-incorporated hema-moetacl hydrogel in rats,”
Scientific Reports (NaturePublisher Group) , vol. 5, p. 9017, 2015.[58] U. Milbreta et al. , “Three-dimensional nanofiber hybrid scaffold di-rects and enhances axonal regeneration after spinal cord injury,”
ACSBiomaterials Science & Engineering , vol. 2, no. 8, pp. 1319–1329,2016.[59] L. H. Nguyen et al. , “Three-dimensional aligned nanofibers-hydrogelscaffold for controlled non-viral drug/gene delivery to direct axonregeneration in spinal cord injury treatment,”
Scientific Reports , vol. 7,p. 42212, 2017.[60] V. Gautam et al. , “Engineering highly interconnected neuronal net-works on nanowire scaffolds,”
Nano letters , vol. 17, no. 6, pp. 3369–3375, 2017.[61] P. Timashev et al. , “3d in vitro platform produced by two-photonpolymerization for the analysis of neural network formation andfunction,”
Biomedical Physics & Engineering Express , vol. 2, no. 3, p.035001, 2016.[62] T. B. DeMarse et al. , “Feed-forward propagation of temporal and rateinformation between cortical populations during coherent activation inengineered in vitro networks,”
Frontiers in neural circuits , vol. 10,p. 32, 2016.[63] A. Gladkov et al. , “Design of cultured neuron networks in vitrowith predefined connectivity using asymmetric microfluidic channels,”
Scientific Reports , vol. 7, no. 1, p. 15625, 2017.[64] P. Assinck et al. , “Cell transplantation therapy for spinal cord injury,”
Nature Neuroscience , vol. 20, no. 05, pp. 637–647, 2017.[65] I. Vismara et al. , “Current options for cell therapy in spinal cord injury,”
Trends in Molecular Medicine , vol. 23, no. 9, pp. 831–849, 2017.66] F.-M. Chen and X. Liu, “Advancing biomaterials of human origin fortissue engineering,”
Progress in polymer science , vol. 53, pp. 86–168,2016.[67] S. Rajangam et al. , “Wireless cortical brain-machine interface forwhole-body navigation in primates,”
Scientific reports , vol. 6, p. 22170,2016.[68] T. Boraud et al. , “From single extracellular unit recording in exper-imental and human parkinsonism to the development of a functionalconcept of the role played by the basal ganglia in motor control,”
Progress in neurobiology , vol. 66, no. 4, pp. 265–283, 2002.[69] J. R. Wolpaw et al. , “Brain–computer interfaces for communicationand control,”
Clinical neurophysiology , vol. 113, no. 6, pp. 767–791,2002.[70] K. Ganguly, “An introduction to brain-machine interfaces.”[71] N. J. Hill et al. , “Recording human electrocorticographic (ecog) signalsfor neuroscientific research and real-time functional cortical mapping,”
JoVE (Journal of Visualized Experiments) , no. 64, pp. e3993–e3993,2012.[72] G. Schalk and E. C. Leuthardt, “Brain-computer interfaces usingelectrocorticographic signals,”
IEEE reviews in biomedical engineering ,vol. 4, pp. 140–154, 2011.[73] W. Wang et al. , “Human motor cortical activity recorded with micro-ecog electrodes, during individual finger movements,” in
Engineeringin Medicine and Biology Society, 2009. EMBC 2009. Annual Interna-tional Conference of the IEEE . IEEE, 2009, pp. 586–589.[74] Z. C. Chao et al. , “Long-term asynchronous decoding of arm motionusing electrocorticographic signals in monkey,”
Frontiers in neuroengi-neering , vol. 3, p. 3, 2010.[75] G. Hotson et al. , “Individual finger control of a modular prosthetic limbusing high-density electrocorticography in a human subject,”
Journalof neural engineering , vol. 13, no. 2, p. 026017, 2016.[76] K. C. Cheung, “Implantable microscale neural interfaces,”
Biomedicalmicrodevices , vol. 9, no. 6, pp. 923–938, 2007.[77] B. Gunasekera et al. , “Intracortical recording interfaces: current chal-lenges to chronic recording function,”
ACS chemical neuroscience ,vol. 6, no. 1, pp. 68–83, 2015.[78] L. F. Nicolas-Alonso and J. Gomez-Gil, “Brain computer interfaces, areview,”
Sensors , vol. 12, no. 2, pp. 1211–1279, 2012.[79] D. J. Krusienski et al. , “An evaluation of autoregressive spectralestimation model order for brain-computer interface applications,” in . IEEE, 2006, pp. 1323–1326.[80] T. J. Cleophas et al. , Machine learning in medicine . Springer, 2013.[81] E. Alpaydin, “Introduction to machine learning (adaptive computationand machine learning series),”
The MIT Press Cambridge , 2004.[82] C. M. Biship, “Pattern recognition and machine learning (informationscience and statistics),” 2007.[83] C. Brunner et al. , “Improved signal processing approaches in anoffline simulation of a hybrid brain–computer interface,”
Journal ofneuroscience methods , vol. 188, no. 1, pp. 165–173, 2010.[84] S. Marsland,
Machine learning: an algorithmic perspective . Chapmanand Hall/CRC, 2011.[85] D. Barber,
Bayesian reasoning and machine learning . CambridgeUniversity Press, 2012.[86] I. Goodfellow et al. , Deep learning . MIT press, 2016.[87] Y. LeCun et al. , “Deep learning,”
Nature , vol. 521, no. 7553, pp. 436–444, 2015.[88] J. Dube, “Fatigue resistibility and stimulus strength using intraspinalmicrostimulation vs. intramuscular stimulation in a rat model: Casestudy,” 2014.[89] P. J. Grahn et al. , “Wireless control of intraspinal microstimulation ina rodent model of paralysis,”
Journal of neurosurgery , vol. 123, no. 1,pp. 232–242, 2015.[90] K. Deisseroth, “Optogenetics,”
Nature methods , vol. 8, no. 1, p. 26,2011.[91] G. Courtine et al. , “Transformation of nonfunctional spinal circuits intofunctional states after the loss of brain input,”
Nature neuroscience ,vol. 12, no. 10, pp. 1333–1342, 2009.[92] N. Wenger et al. , “Closed-loop neuromodulation of spinal sensori-motor circuits controls refined locomotion after complete spinal cordinjury,”
Science translational medicine , vol. 6, no. 255, pp. 255ra133–255ra133, 2014. [93] J. A. Bamford et al. , “The effects of intraspinal microstimulation onspinal cord tissue in the rat,”
Biomaterials , vol. 31, no. 21, pp. 5552–5563, 2010.[94] V. Caggiano et al. , “Rostro-caudal inhibition of hindlimb movementsin the spinal cord of mice,”
PloS one , vol. 9, no. 6, p. e100865, 2014.[95] ——, “An optogenetic demonstration of motor modularity in themammalian spinal cord,”
Scientific reports , vol. 6, p. 35185, 2016.[96] K. L. Montgomery et al. , “Beyond the brain: Optogenetic control inthe spinal cord and peripheral nervous system,”
Science translationalmedicine , vol. 8, no. 337, pp. 337rv5–337rv5, 2016.[97] S. I. Park et al. , “Soft, stretchable, fully implantable miniaturized op-toelectronic systems for wireless optogenetics,”
Nature biotechnology ,vol. 33, no. 12, p. 1280, 2015.[98] K. L. Montgomery et al. , “Wirelessly powered, fully internal optoge-netics for brain, spinal and peripheral circuits in mice,”
Nature methods ,vol. 12, no. 10, pp. 969–974, 2015.[99] M. J. Weber et al. , “A miniaturized ultrasonically powered pro-grammable optogenetic implant stimulator system,” in
BiomedicalWireless Technologies, Networks, and Sensing Systems (BioWireleSS),2016 IEEE Topical Conference on . IEEE, 2016, pp. 12–14.[100] A. Arbabian et al. , “Sound technologies, sound bodies: Medicalimplants with ultrasonic links,”
IEEE Microwave Magazine , vol. 17,no. 12, pp. 39–54, 2016.[101] K. C. Larsson et al. , “Organic bioelectronics for electronic-to-chemicaltranslation in modulation of neuronal signaling and machine-to-braininterfacing,”
Biochimica et Biophysica Acta (BBA)-General Subjects ,vol. 1830, no. 9, pp. 4334–4344, 2013.[102] S. P. Beeby et al. , “Energy harvesting vibration sources for microsys-tems applications,”
Measurement science and technology , vol. 17,no. 12, p. R175, 2006.[103] C. Mead, “Neuromorphic electronic systems,”
Proceedings of the IEEE ,vol. 78, no. 10, pp. 1629–1636, 1990.[104] S. Sa¨ıghi et al. , “Plasticity in memristive devices for spiking neuralnetworks,”
Frontiers in neuroscience , vol. 9, p. 51, 2015.[105] T. Otero et al. , “Biomimetic electrochemistry from conducting poly-mers. a review: artificial muscles, smart membranes, smart drug deliv-ery and computer/neuron interfaces,”
Electrochimica Acta , vol. 84, pp.112–128, 2012.[106] D. T. Simon et al. , “An organic electronic biomimetic neuron en-ables auto-regulated neuromodulation,”
Biosensors and Bioelectronics ,vol. 71, pp. 359–364, 2015.[107] C. Ghez et al. , “Roles of proprioceptive input in the programmingof arm trajectories,” in
Cold spring harbor symposia on quantitativebiology , vol. 55. Cold Spring Harbor Laboratory Press, 1990, pp.837–847.[108] N. G. Hatsopoulos and J. P. Donoghue, “The science of neural interfacesystems,”
Annual review of neuroscience , vol. 32, pp. 249–266, 2009.[109] A. Ramos-Murguialday et al. , “Proprioceptive feedback and braincomputer interface (bci) based neuroprostheses,”
PloS one , vol. 7,no. 10, p. e47048, 2012.[110] G. S. Dhillon and K. W. Horch, “Direct neural sensory feedback andcontrol of a prosthetic arm,”
IEEE transactions on neural systems andrehabilitation engineering , vol. 13, no. 4, pp. 468–472, 2005.[111] K. Horch et al. , “Object discrimination with an artificial hand usingelectrical stimulation of peripheral tactile and proprioceptive pathwayswith intrafascicular electrodes,”
IEEE Transactions on Neural Systemsand Rehabilitation Engineering , vol. 19, no. 5, pp. 483–489, 2011.[112] D. J. Tyler and D. M. Durand, “Functionally selective peripheral nervestimulation with a flat interface nerve electrode,”
IEEE Transactionson Neural Systems and Rehabilitation Engineering , vol. 10, no. 4, pp.294–303, 2002.[113] R. C. Rollings et al. , “Ion selectivity of graphene nanopores,”
Naturecommunications , vol. 7, 2016.[114] D. C. Elias et al. , “Control of graphene’s properties by reversiblehydrogenation: evidence for graphane,”
Science , vol. 323, no. 5914,pp. 610–613, 2009.[115] R. R. Nair et al. , “Fluorographene: A two-dimensional counterpart ofteflon,” small , vol. 6, no. 24, pp. 2877–2884, 2010.[116] J. Zhang et al. , “Communication channel analysis and real timecompressed sensing for high density neural recording devices,”
IEEETransactions on Circuits and Systems , vol. 63, no. 5, pp. 599–608,2016.117] S. Kim et al. , “Thermal impact of an active 3-d microelectrode arrayimplanted in the brain,”
IEEE Transactions on Neural Systems andRehabilitation Engineering , vol. 15, no. 4, pp. 493–501, 2007.[118] K. Abdelhalim et al. , “64-channel uwb wireless neural vector analyzersoc with a closed-loop phase synchrony-triggered neurostimulator,”
IEEE Journal of Solid-State Circuits , vol. 48, no. 10, pp. 2494–2510,Oct 2013.[119] O. Novak et al. , “Wireless ultra-wide-band data link for biomedicalimplants,” in ,July 2009, pp. 352–355.[120] K. M. S. Thotahewa et al. , “Power efficient ultra wide band basedwireless body area networks with narrowband feedback path,”
IEEETransactions on Mobile Computing , vol. 13, no. 8, pp. 1829–1842,Aug 2014.[121] J. Zhang et al. , “Communication channel analysis and real timecompressed sensing for high density neural recording devices,”
IEEETransactions on Circuits and Systems I: Regular Papers , vol. 63, no. 5,pp. 599–608, May 2016.[122] S. Kabiri Ameri et al. , “Graphene electronic tattoo sensors,”
ACS nano ,vol. 11, no. 8, pp. 7634–7641, 2017.[123] S. Lim et al. , “Transparent and stretchable interactive human machineinterface based on patterned graphene heterostructures,”
AdvancedFunctional Materials , vol. 25, no. 3, pp. 375–383, 2015.[124] S. Kim et al. , “Preliminary study of the thermal impact of a microelec-trode array implanted in the brain,” in
Proceedings of the 28th IEEEEMBS Annual International Conference . IEEE, Aug. 2006.[125] K. Ture et al. , “Area and power efficient ultra-wideband transmitterbased on active inductor,”
IEEE Transactions on Circuits and SystemsII: Express Briefs , vol. 65, no. 10, pp. 1325–1329, 2018.[126] K. K. Tokgoz et al. , “A 120gb/s 16qam cmos millimeter-wave wire-less transceiver,” in . IEEE, 2018, pp. 168–170.[127] H. Ando et al. , “Wireless multichannel neural recording with a 128-mbps uwb transmitter for an implantable brain-machine interfaces,”
IEEE Transactions on Biomedical Circuits and Systems , vol. 10, no. 6,pp. 1068–1078, Dec 2016.[128] M. Kang et al. , “Graphene-based three-dimensional capacitive touchsensor for wearable electronics,”
ACS nano , vol. 11, no. 8, pp. 7950–7957, 2017.[129] Q. Zhang et al. , “Human-like sensing and reflexes of graphene-basedfilms,”
Advanced Science , vol. 3, no. 12, 2016.[130] H.-L. C. Kao et al. , “Duoskin: rapidly prototyping on-skin userinterfaces using skin-friendly materials,” in
Proceedings of the 2016ACM International Symposium on Wearable Computers . ACM, 2016,pp. 16–23.[131] X. Huang et al. , “Binder-free highly conductive graphene laminate forlow cost printed radio frequency applications,”
Applied Physics Letters ,vol. 106, no. 20, p. 203105, 2015.[132] D. Seo et al. , “Wireless recording in the peripheral nervous system withultrasonic neural dust,”
Neuron , vol. 91, no. 3, pp. 529–539, 2016.[133] G. E. Santagati and T. Melodia, “Experimental evaluation of impulsiveultrasonic intra-body communications for implantable biomedical de-vices,”
IEEE Transactions on Mobile Computing , vol. 16, no. 2, pp.367–380, 2017.[134] W. Gerstner and W. M. Kistler,
Spiking neuron models: Single neurons,populations, plasticity . Cambridge university press, 2002.[135] F. Xu et al. , “Riemann surfaces of carbon as graphene nanosolenoids,”
Nano letters , vol. 16, no. 1, pp. 34–39, 2015.[136] Y. Tian et al. , “Design and dynamics of a 3-dof flexure-based parallelmechanism for micro/nano manipulation,”
Microelectronic engineering ,vol. 87, no. 2, pp. 230–241, 2010.[137] S. Kralj and D. Makovec, “Magnetic assembly of superparamagneticiron oxide nanoparticle clusters into nanochains and nanobundles,”
ACSnano , vol. 9, no. 10, pp. 9700–9707, 2015.[138] F. Akhtar and M. H. Rehmani, “Energy harvesting for self-sustainablewireless body area networks,”
IT Professional , vol. 19, no. 2, pp. 32–40, 2017.[139] I. Akyildiz et al. , “The internet of bio-nano things,”