BME 4900 Final Report - Biomedical Engineering - University of ...

BME 4900 Final ReportElectronic Circuit to Mimic the Neural Network forthe Saccade ControllerTeam 8Justin Morse, Dean Poulos & Edward RyanProject 12forDr. John EnderleUniversity of ConnecticutDepartment of Biomedical EngineeringA. B. Bronwell Building, Room 209260 Glenbrook RoadStorrs, CT 06269-2247Phone: (860) 486-5521Email: jenderle@bme.uconn.edu

AbstractThe proposed device is an electronic circuit that mimics the neural networkcontrolling fast eye movements, or saccades.The device will simulate the signalsproduced by each neuronal population during the control of a horizontal saccade, andwill allow for observing and recording. It will serve as an extremely valuable teachingtool in the field of neural control, since the timing and synchrony of signals will beprecisely demonstrated. Furthermore, the device will provide possibilities in the realmof diagnosing and properly treating brain injury.Finally, this device could beincorporated into a system for controlling the eye movements of a realistic, artificiallyintelligent robot.The FitzHugh-Nagumo model of the action potential will be used as a foundationto mimic the signals produced by the neurons in question. This is a proven framework,and will provide a precise empirical model that can be customized according to theproperties of a given neuron. The device will be created on printed circuit boards andwill be reasonably priced. This product will be versatile as well as unique, in that acircuit model of this neural network has not been built before.

Table of Contents1 Introduction ............................................................................................................................ 11.1 Background .................................................................................................................................... 11.2 Project Purpose ............................................................................................................................ 21.3 Previous Work Done by Others .............................................................................................. 21.3.1 Products ................................................................................................................................................... 41.3.2 Patents ...................................................................................................................................................... 41.4 Report Outline............................................................................................................................... 42 Project Design ........................................................................................................................ 52.1 Background .................................................................................................................................... 52.2 Optimal Design ............................................................................................................................. 92.2.1 Objective .................................................................................................................................................. 92.2.2 Generalized Neuron Circuit .......................................................................................................... 112.2.3 Superior Colliculus ........................................................................................................................... 192.2.4 Cerebellum ........................................................................................................................................... 202.2.5 Excitatory Burst Neuron ................................................................................................................ 212.2.6 Long-Lead Burst Neuron ................................................................................................................ 212.2.7 Tonic Neuron ...................................................................................................................................... 212.2.8 Omnipause Neuron .......................................................................................................................... 222.2.9 Inhibitory Burst Neuron ................................................................................................................ 222.2.10 Abducens Nucleus .......................................................................................................................... 222.2.11 Oculomotor Nucleus ...................................................................................................................... 232.2.12 Circuitry Case ................................................................................................................................... 232.2.13 Observation of Signals .................................................................................................................. 243 Realistic Constraints ......................................................................................................... 254 Safety Issues ......................................................................................................................... 275 Impact of Engineering Solutions ................................................................................... 286 Life-Long Learning ............................................................................................................. 297 Budget & Timeline ............................................................................................................. 317 .1 Budget .......................................................................................................................................... 317.2 Timeline ....................................................................................................................................... 328 Team Member Contributions ......................................................................................... 358.1 Justin Morse ................................................................................................................................ 358.2 Dean Poulos ................................................................................................................................ 358.3 Edward Ryan ............................................................................................................................... 369 Conclusion ............................................................................................................................ 36

10 References .......................................................................................................................... 3811 Acknowledgements ......................................................................................................... 3912 Appendix ............................................................................................................................. 4012.1 Project Specifications ........................................................................................................... 4012.2 Purchase Requisitions and Price Quotes ....................................................................... 4112.3 Miscellaneous .......................................................................................................................... 42

BME 4900 FINAL REPORT 1Team 81 Introduction1.1 BackgroundDr. John Enderle is a professor of biomedical engineering at the University ofConnecticut. His research focuses in part on the neural network controlling fast eyemovements, or saccades. These movements are performed during such activities asreading and scanning one’s environment.Though the control system behind thesemovements is not completely understood, several parts of the brain are known to have arole. These neuron populations make up a neural network that exhibits coordinatedactivities in the initiation and control of saccades. The model of the network controllinghorizontal saccades is provided in Enderle and Zhou (2010).The research involves investigating this neural network to understand it morefully, and to build a computer model that mimics its behavior. The Hodgkin-Huxleymodel of the action potential is used as a framework. This is an empirical model thatdescribes the behavior of ion channels in the cell membrane that cause potentialchanges.The research is intended to culminate in the development of a way toquantitatively diagnose mild traumatic brain injuries, or concussions.Athletes andconstruction workers are at high risk for this kind of injury, but it can happen to anyone.The effects of multiple, untreated injuries can be additive, leading to a more seriouscondition. In many cases, concussions are difficult to differentiate from normal headpain and dizziness, so the injuries go untreated. The development of a way to definitivelydiagnose the injuries would be a great advancement.This research also has its role in the realm of artificial intelligence. Models of theneural network controlling eye movements can drive the development of robots withrealistic head and eye behavior. The possibilities of such a robot are vast.

2 BME 4900 Final ReportTeam 81.2 Project PurposeThe device will be an electronic circuit that mimics the timing and synchrony ofthe neuronal populations involved in the execution of horizontal saccades. The signalsfrom each neuron center will be observable and recordable. Such a device will be avaluable tool to enhance the understanding of this, and similar, neural networks. Thisdevice could also serve as an input to a robot that must exhibit realistic eye movements.Finally, this product would potentially be a component of a future project that diagnosesmild traumatic brain injuries.This project would culminate in a device that couldobserve a patient’s eye movements and compare them to those of an ideal, uninjuredmodel, to detect the presence of an injury. This product would serve as the referencemodel of the neural network.1.3 Previous Work Done by OthersIn 1952, Alan Hodgkin and Andrew Huxley described an empirical model thatexplains the propagation of action potentials though the behavior of ion channels in thecell membrane. It comprises many differential equations which may be evaluated with anumerical approach, and yield a voltage-versus-time plot of an action potential, whichcan be seen in figure 2. There are parameters that can be changed to yield plots thatapproximate action potentials of neurons with different properties, such as the firing rateand refractory period.Much work has been done since in modeling neuron behavior. In 1961, RichardFitzHugh and J. Nagumo, et al. developed the FitzHugh-Nagumo model, a simplifiedversion of the Hodgkin-Huxley model. It is important because it retains accuracy despiteits simplicity, and it is better suited for implementation in circuitry.More recently, in 1995, a model of the excitatory burst neuron (EBN) was createdby Enderle (Enderle and Zhou, 2010). It was based on the Hodgkin-Huxley model, but

BME 4900 FINAL REPORT 3Team 8with a modified sodium channel equation to achieve a firing rate of about 1000 Hz. TheEBN model also differs from the original model in that it does not require a currentimpulse for a stimulus, but rather a release from inhibition. This model demonstratedthe possibilities of modifying a previous empirical neuron model to simulate any kind ofneuron.In 2006, Dr. Lance Optican described a model of the complete networkcontrolling saccades in Miura and Optican (2006). This work took a different approach,in that it included several more membrane channels, and a different, biochemicallybased scheme for excitation and inhibition of neurons, as opposed to viewing thesesignals as current pulses.The EBN portion of the model sacrificed simplicity forphysiological realism.However, the realism has not been verified by physiologicalexperimentation. The connections between parts of the neural network also differ fromthat proposed by Enderle and Zhou.Recently, Zhou had started a model of the complete saccadic neural networkusing SIMULINK, a simulation tool provided in the MathWorks’ MATLAB suite, and theC++ programming language to ensure a reasonably fast simulation.This model iscreated after Enderle’s vision of the neural network from Enderle and Zhou (2010) andwill serve as the basis for the proposed device. However, the device will use the simplerFitzHugh-Nagumo model for modeling individual neurons.Land (2011) lists several important circuit models of neurons. The model mostimportant to the design of this device is the FitzHugh-Nagumo model, which provides asimplified model of the action potential adapted from Hodgkin and Huxley. A method ofmodeling the synapse is also provided, which can be modified as necessary. Finally,there is an important model that describes the propagation of a pulse in an axon.

4 BME 4900 Final ReportTeam 81.3.1 ProductsElectronic neuron models have been built onto integrated circuits. In one case, asingle neuron was represented on a chip with an area of 4.5 by 5 millimeters (Malmivuoand Plonsey, 1995). With this size, high volumes may be produced, and neural networksmay be created easily. Other neurons with different characteristics have been built basedon existing theoretical models as well. However, these do not seem to be commerciallyavailable, and an integrated circuit provides very little customizability.1.3.2 PatentsThere are no patents for similar devices that will need to be considered whendesigning this product.1.4 Report OutlineThe optimal design of the device will be described in detail.Additionally,discussion of the alternative designs and their shortcomings will be included.Allsubunits of the device will be described, beginning with the parts of each neuron circuitmodel: the dendrite, axon, and synapse. A description of each neuron population andpropagation of signals will follow. Finally, the construction of the device and methods ofdata acquisition will be subsequently described.Constraints due to the environment, sustainability, and manufacturability will bediscussed, as well as safety concerns, the impact of this design on society, and life-longlearning from this project. A description of the budget, timeline, and individual teammember contributions, and a summary will conclude the report.

BME 4900 FINAL REPORT 5Team 82 Project Design2.1 BackgroundThe saccade controller consists of several connected neuron groups that fire insynchrony, based on external feedback, to cause an eye movement. The device willconsist of separate printed circuit boards for each neuron, connected in the mannerpresented in Enderle and Zhou (2010).In all of the alternative designs that weredeveloped, the construction of the device and the modeled neural network remained thesame.The flexibility of design presented itself in the selection of neuron modelsavailable. The three alternative designs under consideration were based on the Harmon,Roy, and FitzHugh-Nagumo models. In considering which model to use, the concernwas with complexity, cost, and accuracy. Safety, environmental, and sustainability issuesdo not differ between models, and also are essentially nonexistent.In 1971, Guy Roy proposed a simple model to reproduce the electrical propertiesof an axonal membrane. The conductance of each is represented by a simple circuitinvolving transistors, resistors, capacitors, and operational amplifiers. The circuits areshown in Fig. 1.Potassium ConductanceSodium ConductanceC6811N9146.8µF2.2kC6815.4k741 7410.12µF56k 39k 326k3.3µF20.4k486k220k39k 0.39µF1.2k14.7k470k560 92.8k12.4k20k2.68k560Figure 1. The potassium and sodium conductance circuits from the Roy model are shown. Supplyvoltages are ± 15 Volts. The output is defined across the source and drain of the transistor ineach.

6 BME 4900 Final ReportTeam 8The field effect transistors (FETs) in this model are made to accurately mimic thetime dependence of actual ion channels. According to the model circuit proposed byHodgkin and Huxley for the capacitive properties of a patch of membrane, theconductance circuits are placed in parallel with a capacitor, and in series with a batterysimulating the resting potential of each ion. This is shown in Fig. 2.50kInsideVInRNa RK RL CMVOutVNa VK VLOutsideFigure 2. The assembled membrane patch circuit is shown. The conductance circuits aresubstituted in the place of R Na and R K. R L is a constant resistance of 220 Ω and C M has a value of0.0047 μF.The results have been compared to data from the experiments of Hodgkin andHuxley on the squid giant axon, and the circuit is shown to be a suitable analog of themembrane. A realistic looking action potential is produced when a current pulse isapplied across the membrane, and voltages are biologically realistic. To be implementedin the proposed neural network, the circuit would need to be modified in order to achievethe firing characteristics given in Enderle and Zhou (2010). This circuit, in comparisonto alternatives, is complicated and no documentation on how to modify itscharacteristics has been found. In the interest of spending less on circuit components,having a simpler design with less room for trouble, and using a circuit that may be easilymodified to match the specifications, this design was rejected.The second alternative design makes use of the circuit proposed by L.D. Harmonin 1961. Figure 3 depicts a simplified version of the Harmon circuit model.

BME 4900 FINAL REPORT 7Team 8-12 VExcitatory Input100k10k 220k 5k0.1µ20k0.1µOutput9.1k 39k 43k 100kInhibitory Input+12 VFigure 3. The preliminary Harmon neuron model circuit schematic. The circuit is not limited tohaving only the specified number of inputs; more can be added as needed.This circuit, using the parameters given above, yields a signal resembling that ofthe Hodgkin-Huxley model of the action potential. The design also allows for explicitdefinition of excitatory and inhibitory inputs, making it significantly easier toaccomodate the multiple input signals of some neuron populations.Using thedocumented properties of the circuit, modifications can be made in order to develop theunique behaviors of the neuron populations being included in the neural network for thesaccade controller. Though modification of this circuit is more feasible than for the Roymodel, it still uses a more than desirable amount of circuit components, and with that, ithas a high level of complexity.Finally, the FitzHugh-Nagumo (FHN) neuron model was also considered. Thismodel is based on the work of Hodgkin and Huxley, and produces similar results with asimpler design. A circuit schematic is found in Fig. 4.

8 BME 4900 Final ReportTeam 80.5 µF2N3904Delayed600Outward100k2N3906CurrentStimulus0.02 mA100kLeakCurrent100k2N3904Fast1kInwardCurrent5 V 0.4 V1 µFFigure 4. A basic FitzHugh-Nagumo model is shown.This design provides a low cost solution, consists of few components, and requireslittle space for construction.One issue is that the output is larger than the truephysiological signal, but this is acceptable according to the specifications, and amplifiercircuits may be used to scale outputs as desired. Adjustments of the leak and outwardcapacitors allow for the circuit to fire at an identical rate to any documented neuralpopulation. The documentation of this design is extensive, the model is very flexible,and matching of the specifications will be feasible.In comparison to the Roy andHarmon models, this circuit is a robust choice and was selected for use in the optimaldesign. Further discussion of the FitzHugh-Nagumo model is in the section of thisreport describing the axon.The neural network, with diagrams indicating relative firing times and rates, anda hypothetical exterior view of the device, are shown in the proceeding section.Additionally, all aspects of the device will be described: the dendrite, the axon, andsynapse of a neuron, the different neuron populations in the neural network, the methodof observing and recording output, and the physical structure of the device. Analyses ofcircuits performed with the National Instruments Multisim circuit simulation suite areincluded.

BME 4900 FINAL REPORT 9Team 82.2 Optimal Design2.2.1 ObjectiveThe objective of this final design is to provide a cheap, cost-effective system thatis capable of mimicking the physiological properties of the horizontal saccade controllerof the brain. In essence, the complete system consists of a series of subsystems designedto imitate the behavior of actual neuronal populations in the horizontal saccadecontroller (see Fig. 5).Figure 5. The neural network for the horizontal saccade generation is shown. Times zero and Trepresent saccade initiation and termination, respectively.

10 BME 4900 Final ReportTeam 8Each of these subsystems is further divided into neural components thatfunction as analogues for different neural structures. Below, in Fig. 6, a diagram of thesecomponents and their interrelations can be found.Action PotentialsFrom Pre-SynapticNeuronsSynaptic Component(Voltage à CurrentConverter)Post-SynapticDendriteComponentPost-Synaptic AxonComponentPost-Synaptic SomaComponentAction Potentials toFollowing Post-Synaptic NeuronsFigure 6. Flow chart describing the propagation and interpretation of signals through theindividual neuron populations.Following this general pattern, each neuron can be modified from a series of“stock” components developed to provide readily available machinery that, as a whole,can provide acceptable descriptions of all neural populations in the horizontal saccadesystem. The first of these components is the synapse, which is responsible for passingthe output of a pre-synaptic neuron to the input mechanisms of the following postsynapticneuron. Here, the voltage action potential signal is converted to a current pulsewhich can be used to excite the following neuron in a specific manner. This currentpulse passes through the dendritic compartments which function as a filter to provide

BME 4900 FINAL REPORT 11Team 8desired input-output relationships. Next, this signal passes to the neural cell soma,which prevents current backflow and primes the axon for excitation. Finally, this signalreaches the post-synaptic axon, which restarts the entire cycle by generating signatureaction potential firing patterns dependent on the type of neural population.2.2.2 Generalized Neuron CircuitDendriteThe dendrite component of the neuron the region that, in most cases, receives thesignal from the synapse of the previous neuron or from some sort of neural receptor. Thesignal is the propagated down the dendrite to the soma. The dendrite is unmyelinated,and as a result, demonstrates significant current leakage. The amount lost is related tothe size of the dendrite, most notably the dendrite diameter and length.The dendrite may be modeled compartmentally using an iterative RC circuit.Each compartment contains three components representing the axial resistance (R a),membrane resistances (R m), and the membrane capacitance (C m) (see Fig. 7).Figure 7. The two possible compartment circuits for a dendrite. The left schematic shows thestandard dendrite compartment, while the right schematic shows the special case of the firstcompartment, which contains one less axial resistance representation.These compartments may be strung in series or in parallel in order to account fordendrite length and any branching that occurs (see Fig. 8). Branching allows for a

12 BME 4900 Final ReportTeam 8neuron to receive multiple stimuli, and is necessary if the neurons are to respond to bothexcitatory and inhibitory inputs.Figure 8. An example of a dendritic tree with multiple branches. The branches converge andterminate at the soma compartment (rightmost subcircuit).In order to accurately model dendrite behavior, the values for the threecomponents must be determined. For this device, it is assumed that the dendrites willhave uniform dimensions and can reasonably be modeled as cylinders. Under theseassumptions, the axial and membrane resistances may be calculated as a function of thecylindrical compartment’s diameter and length (see Eq. 1 and 2).(1) (2)Equations 1 & 2. Determining the medial and axial resistance of a dendrite compartment as afunction of length (l) and diameter (d). Note that R M refers to the specific membrane resistanceand R A refers to the specific axial resistance.Two aspects of the dendrite circuit are being monitored in order to correctlymodel this portion of the neuron: voltage decay and the time constant(s) for thedendrite. The voltage decay is a major factor in determining if the stimulus received willbring the membrane potential at the axon hillock above threshold. The amplitude of theinput current and the simulated dimensions of the dendrite affect the percentage ofvoltage lost. By manipulating the axial resistance and the number of compartments inthe dendrite (largely simulating the length of the dendrite), the voltage decay can bemonitored and analyzed (see Fig. 9). The data gathered shows that as the axial resistance

BME 4900 FINAL REPORT 13Team 8decreases, the linearity of the voltage decay curve increases and the rate of decaydecreases.Figure 9. Percent voltage loss as a function of dendrite length. Each curve denotes a different axialresistance.The other aspect of the dendrite that must be manipulated is the time constant ofthe RC circuits. In addition to the voltage decay, this value, describing the amount oftime it takes for the capacitors to charge. If the time constant is too large, then it ispossible that the membrane potential will not reach threshold in the desired time,drastically altering the behavior of the neuron. Theoretical calculation of the equivalenttime constant for the complete dendrite circuit is possible, but is difficult to interpret asan analytical equation. Fortunately, developing a program to perform the iterativecalculations allows for rapid determination of the time constant if the component valuesare known for the system.Soma (Cell Body)The soma acts, functionally, as the final compartment of the dendritic tree,consolidating all of the input signals. Unlike the other dendrite compartments, the somamodel contains no axial resistance element. However, there is a new property that must

14 BME 4900 Final ReportTeam 8be incorporated: a current stop to prevent backflow into the dendrite from the axon. Thediode would typically be used to create such a stop, but due to the voltage loss over thediode, an alternative method is used. The precision diode operational amplifier circuithelps to eliminate the voltage drop over the diode while still allowing for the backflow tobe stopped. It is followed by a voltage follower operational amplifier circuit (see Fig. 10).Multisim analysis of the compartment’s circuit shows a current backflow ofapproximately 53 nA, which is an acceptable backflow given the amplitudes we aredealing with in the model.Figure 10. A generalized schematic of the soma compartment, including the current backflowstop. This circuit outputs to the axon compartment.AxonThe axon is final component of the neural unit and, consequently, the site ofaction potential generation and propagation. The pre-synaptic input, which crosses thepreviously discussed synapse and is conducted by the post-synaptic dendrites, finallyreaches the neuraon axon where a resulting neural action is produced. In the context ofour final design, several basic subcircuit designs can be repeated, with modification, tomimic the desired behavior of each neuron population. The design relies on theFitzHugh-Nagumo circuit model of a neuron, which is an adaptation of the empiricallydefined Hodgkin-Huxley model. This analog design provides a robust, cost-effective

BME 4900 FINAL REPORT 15Team 8solution for the range of behaviors exhibited by each neuron population of interest. Asshown in Fig. 11 below, the “stock” design of this circuit is relatively simple and providesa train of action potentials at approximately 1000 Hz.Figure 11. The circuit schematic for “stock” 1000 Hz FitzHugh-Nagumo axon.12 below.The behavior of this model is displayed in the NI Multisim trace available in Fig.Figure 12. The NI Multisim trace for 1000 Hz FitzHugh-Nagumo axon.While this “stock” model functions properly for autonomously firing 1000 Hzneurons, some additional work is necessary to describe the desired physiologicallyaccurate neuronal populations. To encompass all of the desired neural behaviors of eachof these various populations, several modifications are necessary which directly impactthe firing rate. These modifications, while relatively simple in nature, allow for profound

16 BME 4900 Final ReportTeam 8changes in firing rate and input current range.Concisely, these changes involvemodification of 15 nF leak current channel capacitor which, in turn, allows for a variablefiring rate between 200 – 1300 Hz, depending on the input current pulse. A morecomprehensive description of this behavior can be found in Fig. 13 below, which displaysthe novel firing rate versus input current for different leak channel capacitances.Figure 13. Firing Rate vs. Input Current for Variable Leak Channel CapacitanceIn the end, these modifications allow us to easily generate axon components thatare capable of mimicking all the desired behaviors of each neural population. Table 1contains the pertinent frequency data that describes each of the neuron populationsbeing modeled.

BME 4900 FINAL REPORT 17Team 8Neural SiteOnset BeforeSaccadePeak Firing RateApproximate EndTimeAbducens Nucleus 5 ms 400-800 Hz 5 ms before saccadeendsContralateralSuperior Colliculus(SC)Ipsilateral ExcitatoryBurst Neurons (EBN)Ipsilateral InhibitoryBurst Neurons (IBN)Ipsilateral Long-LeadBurst Neurons(LLBN)20-25 ms 800-1000 Hz At saccadetermination6-8 ms 600-800 Hz 10 ms before saccadeends6-8 ms 600-800 Hz 10 ms before saccadeends20 ms 800-1000 Hz At saccadeterminationOmnipause Neurons(OPN)6-8 ms 150-200 Hz(before and after)At saccadeterminationTable 1: The neuron populations being modeled and the associated timings and frequencies.SynapseThe synapse is the chemical or electrical connection between two neurons. In thesynaptic terminal, which is the part of the neuron following the axon, action potentialfrequency causes the proportional release of neurotransmitter. In a circuit model, thisrelease may be modeled as a current pulse. To obtain similar behavior in this model, twocircuits are used: the tachometer and bilateral current source. The tachometer is shownin Fig. 14.

18 BME 4900 Final ReportTeam 8Figure 14. The tachometer circuit, using both halves of a LM13700 operational amplifier chip, isshown. The voltage at the output has a linear relationship with the frequency of the input.The left part of the circuit in figure T is a comparator, and the reference voltage isset at 1V because it is a point that the action potentials reliably cross each time and willprovide a good frequency measure. The voltage across R1 builds and maintains a valueproportional to the frequency of the input by the relationship, .The bilateral current source is shown in Fig. 15.Figure 15. The bilateral current source circuit, using a LM108AD precision operational amplifier,is shown. The current at the output is constant regardless of load and has a linear relationship tothe voltage at the input.

BME 4900 FINAL REPORT 19Team 8The input will be connected to the output of the tachometer. The current outputis related to the voltage at the input by the relationship in Equation 3.Equation 3. Determining the output current of the bilateral current source. Note that for thisequation, and .The current output will be injected into the dendrite of the post-synaptic neuronand will either stimulate or inhibit it. The inhibitory or stimulatory nature of the synapseis dictated by the addition of either of the following circuits mentioned in Land (2011).The circuits are shown in Fig. 16.Figure 16. The excitatory and inhibitory circuits are shown. The outlined resistors may be varied to increaseor decrease synaptic strength.2.2.3 Superior ColliculusThe superior colliculus receives information from other portions of the brainabout how far the eye should move. It outputs a signal whose length is proportional tothe magnitude of the desired movement and initiates the action of the rest of theneurons. This part of the device will be created in conjunction with the eye-movementrobot design team, who will be designing the image processing software.

20 BME 4900 Final ReportTeam 8Information from the robot’s cameras will be processed and the distance anddirection the eyes must move will be supplied by the robot. The appropriate half of thesuperior colliculus will fire for a period of time related to this information. The cameraswill continuously monitor the LED array before it. When a new LED is lit, the robot willshift its gaze to it.Fig. 17 outlines the procedure for mimicking the superior colliculus.EyeMovementRobotImage FeedImage Processingto Locate LEDSCalculations toDetermine Changein StateAngle andDirection ofDesiredMovementGeneration ofMovement viaNeural NetworkSignalImitatingSuperiorColliculusGeneration ofProportionalSignalFigure 17. A flowchart illustrating the operation of the superior colliculus is shown. The processis initiated with information provided by the robot, indicated in red. The first four steps will bepart of the robot team’s design and the last three will be implemented in this design.Before the robot-neural network integration takes place, a signal representingsaccade direction and magnitude will be generated with LabVIEW and used as the inputto the neural network. Thus, this device will be able to operate independently of an eyemovementrobot and be used as a model at any time.2.2.4 CerebellumAt the request of the client, the cerebellum elements of the neural network,notably the fastigial nucleus, will not be included in the circuit design. The superiorcolliculus will function such that it will fulfill the role of the input to the neuralpopulations that would otherwise be connected to the fastigial nucleus. This substitutionwill be referred to when describing the inputs and outputs of the other neuronpopulations.

BME 4900 FINAL REPORT 21Team 82.2.5 Excitatory Burst NeuronThe excitatory burst neuron, located in the paramedian pontine reticular formation,serves as one of the major excitatory inputs for the saccade controller. Firing at a rate ofapproximately 1000 Hz, this neuron fires spontaneously upon release from inhibition.The primary inputs for this neuron population, based on the model being simulated, arethe excitatory input of the superior colliculus and the inhibitory inputs of the inhibitoryburst and omnipause neurons. The circuit design for this neuron population will employthe FitzHugh-Nagumo burster model attached to both an excitatory and inhibitorysynapse. These synapses will provide excitatory inputs to the tonic neurons and theabducens nucleus, and inhibit the inhibitory burst neuron during firing.2.2.6 Long-Lead Burst NeuronThe long-lead burst neuron shares a location and similar function to theexcitatory burst neuron, but is instead responsible for controlling the behavior of theomnipause and inhibitory burst neurons.This neuron population, in the proposedmodel, will be controlled exclusively by the superior colliculus.The circuitrepresentation of the long-lead burst neuron will contain a FitzHugh-Nagumo burstermodel with both an excitatory and an inhibitory synapse, with the excitatory signal beingsent to the inhibitory burst neuron and the inhibitory signal being relayed to theomnipause neuron.2.2.7 Tonic NeuronThe tonic neurons are responsible for fixing the rectus eye muscles in place once thesaccade completes. This neuron population receives excitatory and inhibitory inputsfrom the excitatory and inhibitory burst neurons, respectively. The tonic neurons will be

22 BME 4900 Final ReportTeam 8represented in the circuit system by a FitzHugh-Nagumo burster model with anexcitatory synapse, which will transmit the appropriate signal to the abducens nucleus.2.2.8 Omnipause NeuronThe omnipause neuron serves as an inhibitory signal to keep the neural networkat rest in between saccades. It receives exclusively inhibitory inputs from both of thelong lead burst neuron groups (one on either side of the system).This neuronpopulation also provides only inhibitory outputs, one going to each of the inhibitoryburst neuron groups. The omnipause neuron will be represented electronically by aFitzHugh-Nagumo burster model with an inhibitory synapse that will deliver theappropriate signal to the inhibitory burst neuron circuits.2.2.9 Inhibitory Burst NeuronThe inhibitory burst neuron controls the firing of the excitatory burst neuron aswell as the tonic neuron, both of which are on the opposite side of the system to thecorresponding inhibitory burst neuron.This neuron population receives excitatoryinput, in this model, from the superior colliculus and the long-lead burst neuron, and aninhibitory input from the omnipause neuron.The inhibitory burst neuron will beimplemented via the use of the FitzHugh-Nagumo burster model with an inhibitorysynapse in order to provide the proper inputs to the system.2.2.10 Abducens NucleusThe abducens nucleus functions as the input for the lateral rectus eye muscles,while also influencing the behavior of the oculomotor nucleus of the opposite side. Theabducens nucleus is excited by the excitatory burst neuron during the saccade and by thetonic neuron once the saccade has completed. The inhibitory burst neuron inhibits this

BME 4900 FINAL REPORT 23Team 8portion of the system outside of the saccade execution period. The circuit model for theabducens nucleus will contain the FitzHugh Nagumo burster model with an excitatorysynapse.2.2.11 Oculomotor NucleusThe oculomotor nucleus is solely responsible for the control of the medial rectuseye muscles. This nucleus receives excitatory input from the abducens nucleus andinhibitory input from the inhibitory burst neuron. The circuit implementation of thiselement will include a FitzHugh-Nagumo burster model and a single excitatory synapse,giving the output to the muscle analog.2.2.12 Circuitry CaseIn order to allow for feasible movement and management of the neural networkcircuitry, the circuit boards will be connected inside a case.However, due to thecurrently unknown size of the circuit boards being produced, the actual dimensions ofthis enclosure have yet to be determined. The case will be made of opaque acrylic,allowing for a clean finish and easy manufacturing. The structure will be reinforced withaluminum angle to provide additional structural integrity. There will also be openingsallowing for assisted ventilation from the cooling system as well as access to the circuitboards themselves.Upon construction, the case will have locations for the user to connect leads toobserve the action potentials developed by each of the neuron populations in the circuit,including the final outputs for the medial and lateral rectus eye muscles. The user mayconnect to as many or as few leads as desired, allowing for selective analysis of thesystem.

24 BME 4900 Final ReportTeam 82.2.13 Observation of SignalsThere will be 17 neuron groups represented in the device, and the output signalsof all of them will be observable. The contacts on the case may be observed with anoscilloscope or connected to NI Data Acquisition (DAQ) hardware so the signal may beprocessed in LabVIEW. With the NI hardware at hand, eight analog inputs are available.Thus, there will be eight “channels” which will be able to record any of the neuronoutputs simultaneously. Fig. 18 illustrates the data acquisition process for this device.AnalogSignal fromNeuronCircuitNI DAQAcquisition DeviceSoftware FiltersVoltage v.Time Graphof SignalSpike Counting VIand CalculationsFiring RateDisplayFigure 18. A flowchart for the data acquisition process is shown.The signals from multiple neurons may be acquired with the NI DAQ hardware.Once the signal has been acquired by LabVIEW, software filters will be applied asnecessary to eliminate noise from the signals. The treated signals will be displayed on agraph with an appropriate time and voltage scale. A spike counting virtual instrument(VI) will be used to calculate the firing rate of the signal, and it will be displayed on thefront panel.

3 Realistic ConstraintsBME 4900 FINAL REPORT 25Team 8Due to the size and application of this neural network, there are no organizationsthat must approve the manner in which the project is performed. The circuit is not to beimplanted in an individual, and therefore does not need to be approved by the FDA. Thefinal product will also not be of such a size that structural or mechanical issues pose aserious concern, and require no certification in that regard. The constraints on theproject are subtle, but do limit its functionality in some aspects.The design of the circuit model is such that it represents the major populations ofneurons involved in producing the signals for the lateral and medial rectus models,yielding a certain degree of physiological realism. The resulting signals do resemble theactual action potential with regards to their firing rates, though the amplitudesimmediately produced are not as accurate. The functionality of the circuit has beengiven priority of the physiological realism of the amplitude. The measurable signals canbe dampened in order to yield more appropriate amplitudes, but the fact still remainsthat the circuits, on their own, do not yield the expected physiological voltages.The circuits for the individual neuron populations are designed such that theycannot be altered once connected to the printed circuit board. This property means thatif additional information becomes available that would suggest altering the behavior ofneuron, the circuit board is more likely to need to be completely replaced. However, thisdesign choice gives more reliable, durable neuron circuits. Allowing for interchangeableparts would result in the inability to solder components into place, greatly increasing thepossibility of parts becoming loose during handling, causing the entire neural network toyield inaccurate signals.Beyond these implementation constraints, the device does not create any sort ofcontroversy with regards to its production. Of all the proposed designs, this methodinvolved the least quantity of parts, ultimately yielding the least expensive model. This

26 BME 4900 Final ReportTeam 8will allow for the project to be completed with a lower budget, as less circuit componentsand circuit board space would be required.No controversy is expected to arise with regards to the device itself either. Beingcompletely comprised of circuitry, there is no need for any sort of in vitro or in vivotesting, meaning that no animals or cells need be harmed in order to develop theproduct. It is true that animal testing has been performed in order to obtain some of thephysiological data used to estimate the parameters that the device is based on; noadditional testing is needed in this form. The device also is not meant to alter a humanor animal in any fashion, so ethical concerns related to this are expected to benonexistent.The device, if used properly, should not be difficult to maintain. The circuitelements remain static on the appropriate circuit boards, and should not come looseduring regular use. The connections between boards should also remain connected, asthe cables contain clips within their structures that encourage the connections to remaintight when placed in the appropriate receptacle. The circuits and connections should beexamined, however, in the event of the case being dropped.

4 Safety IssuesBME 4900 FINAL REPORT 27Team 8This design, due to its extensive use of electronic components, requires properhandling of two major safety concerns: electrical and thermal.The circuits beingdesigned require specific voltage levels in order to function, but these all occur at orbelow five volts. Harm to the operator related to electrical, if any, would likely be theresult of misuse of the circuitry, resulting in minor electrical shock.The signalsdeveloped by the model will also be processed by a computer, which brings about its ownsafety concerns. The safety issues to be addressed in this regard, however, are largelydependent upon the model computer being used.Operators should consult themanual(s) for that device in order to ensure they are following safety protocol.Generalized safety issues would generally be the result of connecting the wrong leads forsignal transmission, which again would possibly lead to minor electrical shock. Moreserious injury could result if the operator decides to manipulate the computer partsduring use, though this is in no way required or recommended when using the device.Thermodynamics dictate that during the use of electric circuit components, heatis generated.With numerous circuits running simultaneously, the amount of heatgenerated increases significantly. The operator should not have to touch any of thecircuit components during operation, though if this were to occur, any injury would likebe seen in the form of first degree burns. Any further injuries (more serious burns)would suggest severe misuse of the product. In order to minimize the possibility of thisoccurring, the container for the circuitry will include a fan (or multiple fans if necessary)in order to keep the parts cool, avoiding operator injury and failure of circuit elements.

28 BME 4900 Final ReportTeam 85 Impact of Engineering SolutionsAs a whole, the development of an analog electronic neuron does have somepotential implications that can be discussed.The proposed device allows for theconstruction of a physiologically accurate eye saccade control system. Assuming that theaccompanying muscular system can be developed elsewhere, these models combinedwould provide a complete functioning eye control system that could be exported andextrapolated to other robotic designs with minimal change being required.This isconvenient for any biomimetic system that requires eye control and motion. However,the benefits of this design are not limited to strictly robotic applications.Because the system is physiologically accurate, a complete robotic human eyeanalog can be used to diagnose mild traumatic brain injuries, often referred to asconcussions.When an individual suffers a mild traumatic brain injury, there aregenerally few or no symptoms of any brain damage that may be noticed qualitativelyduring examination. Using this device as an input benchmark, the neural signals andresulting eye motions may be compared to that of a physiologically “normal” saccade.Deviation from this control can suggest the extent of brain damage for the patient,allowing for early diagnosis and treatment. This early action can help to avoid long-termpain, brain-related illness, and possibly even death due to injuries sustained during themild traumatic brain injury.All of these applications entail realistic prospect for the finished device.Acomplete, easy-to-manage saccadic control system can revolutionize modern robotics.On the opposite end of the spectrum, a complete system can also be used in medicine toaid in the diagnosis of traumatic brain injury.

6 Life-Long LearningBME 4900 FINAL REPORT 29Team 8Brain physiology and anatomy, the empirical model, control systems, the art ofcircuit building and troubleshooting, and the design process are all topics in whichknowledge will be and has been acquired due to the creation of this device. These arecritical pieces of information and skills in engineering, and they will continue to bevaluable.The complexity of the neural network for the control of such a simple task, thehorizontal saccade, is staggering. An appreciation of the beauty of this naturally evolvedcontrol system can be gained from this project. Even the fastest computers cannotoutperform the brain. An understanding of the brain’s systems, and thinking about howman-made systems can more closely mirror them, could lead to improved efficiency andpower in systems.In the neural network, the difference between desired eye location and actual eyelocation is encoded in a signal sent to the superior colliculus, and the amount andlocation of neurons firing there initiate a chain of relayed signals to the rest of thenetwork. The populations on each side of the midline excite and inhibit each otherappropriately to guarantee movement between the eyes is coordinated. The error isconstantly fed back to the superior colliculus, ensuring the proper outcome is reached.This kind of scheme is used universally in control systems, and is vital to understand.On a lower level, simply an understanding of brain physiology and anatomy, suchas the neuron populations involved in various tasks, the nature of membrane potentials,and the behavior of ion channels, is a useful thing. The value of the empirical model isbrought to light in this project. The model is not an analog to the actual physiologicalprocess, in that it does not replicate the behavior of every component of the real system.However, it provides results that match the outcome of the process. The Hodgkin-Huxley model, the basis for the simpler FitzHugh-Nagumo model, is an apt example,

30 BME 4900 Final ReportTeam 8describing the electrical behavior of the neuron membrane during an action potentialusing differential equations. It was built by matching experimental data from a squidaxon, not by building a replica of the axon in an attempt to make it behave the same way.The Hodgkin-Huxley model remains an extremely important contribution, anddemonstrates the importance of applying empirical models.The device will useempirical circuit models of neurons that are not physiologically analogous to realneurons, but perform analogous actions.Circuit design is a meticulous process because all aspects of the circuit must beabsolutely correct and when malfunction occurs, it is often difficult and frustrating tofind the cause. The building of this device will be a lesson in proper technique forcreating and troubleshooting complicated circuits. Complementary to this is the designprocess in general.Proper documentation of steps, planning and budgeting time,money, and resources are necessary in a successful project. The device will have athorough owner’s manual, the progress and design are documented in periodic reportsand presentations, and time and resources will be tracked in a Microsoft Project file. Theprocess of building this device mirrors the engineering process in industry and willprovide valuable life-long skills.

BME 4900 FINAL REPORT 31Team 87 Budget & Timeline7 .1 BudgetAt the moment, the current budget outlook remains highly fluid. The currentproposal utilizes basic electrical circuit components that are readily available and low incost. The specifics of the current design remain a topic of discussion and, as a result, thebest estimate can be derived from the previous constructions of similar products.Malmivuo and Plonsey discuss three different electronic neuron models in their article,each using a similar quantity of circuit components. In order to construct any of theproposed models, the circuits must contain resistors, capacitors, transistors, and diodes.From the relative quantities in each of the models, it would cost roughly $20 in circuitelements per neuron population. This would give a total cost of $320 in parts.The largest expenditure for the project would be purchasing the circuit board(s)to layout the design. The current estimate places this cost at approximately $300 total,and is subject to change based on layout decisions. An enclosure for the circuitry,comprised of an aluminum skeleton with acrylic walls, would cost an additional $50.Several portions of the project will be available at no cost. Access to machines tomanufacture the enclosure is available, as is the LabVIEW software that may be used todevelop the signal analysis program. Input sources may also be available at no cost.These estimates lead to an exact total of $670. The requested budget for theproject is $700, allowing for some flexibility in the estimates provided.Table 2: The breakdown of the items purchased for the project thus far.

32 BME 4900 Final ReportTeam 87.2 TimelineThis table is a summary of all the specified tasks for the project. A Microsoft Projectversion of this data may be found on the projects website at:.Name Duration Start Finish PredecessorsAnalyze FHN circuit parameters 1 wk 10/17/11 10/21/11Research case manufacturers 1 wk 10/17/11 10/21/11Analyze excitatory synapse circuit parameters 1 day 10/24/11 10/24/11Analyze inhibitory synapse circuit parameters 1 day 10/26/11 10/26/11Analyze dendrite circuit behavior 2 wks 10/27/11 11/9/11Test excitatory synapse connection in Multisim 1 day 10/31/11 10/31/11 1,2Test inhibitory synapse connection in Multisim 1 day 11/1/11 11/1/11 1,3Design LLBN axon hillock 1 day 11/10/11 11/10/11 1Design EBN axon hillock 1 day 11/10/11 11/10/11 1Design TN synapses 1 day 11/10/11 11/10/11 15,2Design LLBN synapses 1 day 11/11/11 11/11/11 3,5,2Design EBN synapses 1 day 11/11/11 11/11/11 3,10,2Design LLBN dendrites 1 day 11/14/11 11/14/11 6,4Design EBN dendrites 1 day 11/14/11 11/14/11 11,4Design TN dendrites 1 day 11/14/11 11/14/11 16,4Design LLBN axon timing circuit 1 day 11/15/11 11/15/11 7Design EBN axon timing circuit 1 day 11/15/11 11/15/11 12Design TN axon timing circuit 1 day 11/15/11 11/15/11 17Assemble LLBN in Multisim 1 day 11/16/11 11/16/11 5,6,7,8Assemble EBN in Multisim 1 day 11/16/11 11/16/11 10,11,12,13Assemble TN in Multisim 1 day 11/16/11 11/16/11 15,16,17,18Design OPN axon hillock 1 day 11/17/11 11/17/11 1Design IBN axon hillock 1 day 11/17/11 11/17/11 1Design AN axon hillock 1 day 11/17/11 11/17/11 1Design OPN synapses 1 day 11/18/11 11/18/11 20,3Design IBN synapses 1 day 11/18/11 11/18/11 25,3Design AN synapses 1 day 11/18/11 11/18/11 30,2Test neuron pair where both syanpses exist inMultisim2 days 11/2/11 11/3/11 1,2,3Design OPN dendrites 1 day 11/28/11 11/28/11 21,4

BME 4900 FINAL REPORT 33Team 8Design IBN dendrites 1 day 11/28/11 11/28/11 26,4Design AN dendrites 1 day 11/28/11 11/28/11 31,4Design OPN axon timing circuit 1 day 11/29/11 11/29/11 22Design IBN axon timing circuit 1 day 11/29/11 11/29/11 27Design AN axon timing circuit 1 day 11/29/11 11/29/11 32Assemble OPN in Multisim 1 day 11/30/11 11/30/11 20,21,22,23Assemble IBN in Multisim 1 day 11/30/11 11/30/11 25,26,27,28Assemble AN in Multisim 1 day 11/30/11 11/30/11 30,31,32,33Design TN axon hillock 3 days 11/7/11 11/9/11 1Design OCN axon hillock 1 day 12/1/11 12/1/11 1Design power distribution circuit 1 wk 12/1/11 12/7/11 1,2,3,4Design OCN synapses 1 day 12/2/11 12/2/11 35,2Design OCN dendrites 1 day 12/5/11 12/5/11 36Design OCN axon timing circuit 1 day 12/6/11 12/6/11 37Assemble OCN in Multisim 1 day 12/7/11 12/7/11 35,36,37,38Assemble Neural Network in Multisim 1 day 12/8/11 12/8/119,14,19,24,29,34,39,40,44Design power distribution PCB 1 day 12/8/11 12/8/11 40Design generic PCB 1 day 12/9/11 12/9/11 41Design attenuator for intermediate output signals 1 day 12/9/11 12/9/11 41Program LabVIEW signal acquisition program 2 wks 12/19/11 12/30/11 1,2,3,4Design SC LabVIEW program (user interface) 2 wks 12/19/11 12/30/11Program LabVIEW signal processing program 2 wks 1/2/12 1/13/12 105Design SC LabVIEW program to convert robotsignal to saccade size2 wks 1/2/12 1/13/12 107Test LLBN axon on Protoboard 1 day 1/16/12 1/16/12 5Test EBN axon on Protoboard 1 day 1/16/12 1/16/12 10Test TN axon on Protoboard 1 day 1/16/12 1/16/12 15Test LLBN axon + synapse on Protoboard 1 day 1/17/12 1/17/12 6,57Test EBN axon + synapse on Protoboard 1 day 1/17/12 1/17/12 11,60Test TN axon + synapse on Protoboard 1 day 1/17/12 1/17/12 16,63Test full LLBN on Protoboard 1 day 1/18/12 1/18/12 9,58Test full EBN on Protoboard 1 day 1/18/12 1/18/12 14,61Test full TN on Protoboard 1 day 1/18/12 1/18/12 19,64Test OPN axon on Protoboard 1 day 1/19/12 1/19/12 20Test IBN axon on Protoboard 1 day 1/19/12 1/19/12 25

34 BME 4900 Final ReportTeam 8Test AN axon on Protoboard 1 day 1/19/12 1/19/12 30Test OPN axon + synapse on Protoboard 1 day 1/20/12 1/20/12 21,66Test IBN axon + synapse on Protoboard 1 day 1/20/12 1/20/12 26,69Test AN axon + synapse on Protoboard 1 day 1/20/12 1/20/12 31,72Test full OPN on Protoboard 1 day 1/23/12 1/23/12 24,67Test full IBN on Protoboard 1 day 1/23/12 1/23/12 29,70Test full AN on Protoboard 1 day 1/23/12 1/23/12 34,73Test OCN axon on Protoboard 1 day 1/24/12 1/24/12 35Test excitatory synapse on Protoboard 1 day 1/24/12 1/24/12 59,68Test power distribution circuit on Protoboard 2 days 1/24/12 1/25/12Test OCN axon + synapse on Protoboard 1 day 1/25/12 1/25/12 36,75Test inhibitory synapse on protoboard 1 day 1/25/12 1/25/12 59,71Test full OCN on Protoboard 1 day 1/26/12 1/26/12 39,76Test neuron pair where both syanpses exist onProtoboard2 days 1/26/12 1/27/12 78,79Design signal filters as necessary 1 day 1/27/12 1/27/1259,62,65,68,71,74,77Design LLBN PCB 1 day 1/30/12 1/30/12 45,9Design EBN PCB 1 day 1/30/12 1/30/12 45,14Design TN PCB 1 day 1/30/12 1/30/12 45,19Design OPN PCB 1 day 1/31/12 1/31/12 45,24Design IBN PCB 1 day 1/31/12 1/31/12 45,29Design AN PCB 1 day 1/31/12 1/31/12 45,34

BME 4900 FINAL REPORT 35Team 88 Team Member Contributions8.1 Justin MorseJustin has done much exploratory work, and has found the existing neuronmodels from which the project is being based. His contribution to the documentation isequal to that of the other team members, and has taken part in upkeep of the website.He also modified a FitzHugh-Nagumo model to fire at 1000 Hz, both with and without astimulatory current input, and these are currently being used for axons. Though hisearly work focused on characterizing the dendrite, Ed has taken these reigns and Justinhas become the leader of the work on the synapse, which emerged more recently as animportant issue.His work on the dendrite included developing the scheme to replicate truedendrite behavior using analog circuits, investigating possible circuits for use in thesynapse, and customizing the circuits for use in the project. Further work will involvemore faithfully replicating the results found in GENESIS simulations, combining thesynapse with an axon and dendrite to create one functional neuron.8.2 Dean PoulosMost of Dean Poulos’ early work revolved around the construction of anappropriate axon unit. At first, this entailed heavy research into the previous works ofother academic scholars who have focused on mimicking the physiological behavior ofbasic neurons using analog components. Once a viable model was found, further workpursued regarding the perfection of this behavior and providing readily availablealternatives that could be utilized for the construction of all neural populations in thesystem.This was achieved by weeks of testing and experimentation to further theunderstanding of the workings of these axon units, primarily focusing on how these

36 BME 4900 Final ReportTeam 8variable components could be modified to mimic the desired firing rates for eachpopulation.In the end, this work allowed for a more complete understanding of theFitzHugh-Nagumo axon unit. Three primary classes of neurons have been based uponthis design, allowing for easy preparation of custom axon units. In addition, Dean alsoplayed a major role in the development of the primitive synaptic units, prior to thechanges installed by Justin and Ed later in the year. Again, this portion of the projectfocused on understanding how these pre-built synapse units functioned, what controlledtheir behavior and how this behavior could be modified and prepared in a readilyavailable manner.8.3 Edward RyanEdward has allocated most of his time in two areas of the project: analysis ofdendrite behavior and building the team’s infrastructure. Very little information hasbeen published with regards to the implementation of dendrite models, and as such,most of the information to use as a framework had to be either found experimentally orgathered from the equations provided in The Book of Genesis section on dendrites. Thetwo properties of interest in the dendrite are the voltage decay and the time constant thatis created as a result of the RC model being used. Analytical interpretations of the systemproved to be difficult to manipulate, therefore two alternative approaches were taken.The circuit was constructed in NI Multisim, where the voltage decay over the dendritebased on its component parameters and number of compartments could be determined.As for the time constant, a short MATLAB function was created in order to allow for theuser to input the component values and the desired dendrite length (in compartments),and the appropriate time constant could be calculated in that manner.

BME 4900 FINAL REPORT 37Team 8The team’s infrastructure has been supported by all members of the group, withEdward having set up a large portion of the framework. The website template providedfor the group’s use was expanded on to provide a more viewer-friendly and professionaldesign. Information sharing methods such as Dropbox folders and other arrangementswere also prepared so that the group could improve its communication, particularly overperiods where the group may no be able to meet in person for extended periods of time.9 ConclusionThis device will use an electronic circuit to mimic the neural network controllinghorizontal fast eye movements, or saccades. The signals produced by every neuronpopulation involved will be observable and recordable. The product will incorporateprevious work on neuron models into a neural network that has not been represented inthis manner before. It will provide an enhanced understanding of this neural networkand will be a stepping stone for other projects, such as a the control of a robot’s eyemovements, and diagnosing mild traumatic brain injuries. Additionally, the device willuse traditional analog circuit components and repeated design elements, keeping itaffordable. The possibilities this device holds for the fields of artificial intelligence andneural medicine are great, and the creation of this product will be a great step forward inthe field of neural modeling.

38 BME 4900 Final ReportTeam 810 ReferencesEnderle, J.D., and Zhou, W., Models of Horizontal Eye Movements. Part 2: A 3rd-OrderLinear Saccade Model, Morgan & Claypool Publishers, San Rafael, CA, 2010, 144 pagesMalmivuo, J., & Plonsey, R. (1995). Bioelectromagnetism. (1 ed., pp. 168-80). NewYork: Oxford University Press.Retrieved from http://www.bem.fi/book/10/10.htmMiura K. and Optican L.M.: Membrane channel properties of premotor excitatory burstneurons may underlie saccade slowing after lesions of omnipause neurons. J ComputNeurosci, 20: 25-41, 2006.J. M. Bower and D. Beeman, The Book of GENESIS – Exploring Realistic Neural Modelswith the GEneral NEural SImulation System, Internet Edition: ,2003.Land, B. R. (1997). Analog and digital hardware neural models. Informally publishedmanuscript, Department of Neurobiology, Cornell University, Ithaca, New York.Retrieved from http://www.nbb.cornell.edu/neurobio/land/PROJECTS/NeuralModels/index.htmlHarmon LD (1961): Studies with artificial neurons, I: Properties and functions of anartificial neuron. Kybernetik Heft 3(Dez.): 89-101.Roy, Guy, A simple electronic analog of the squid axon membrane, IEEE Trans BiomedEng. 1972 Jan; 19(1): 60-3.

BME 4900 FINAL REPORT 39Team 811 AcknowledgementsThere are a number of individuals who have been necessary for this project’snatural progression and eventual completion. The first, and perhaps most obvious ofthese, is Dr. John Enderle of the University of Connecticut’s Biomedical EngineeringDepartment.Doubling as both our project’s sponsor and mentor, Dr. Enderle hasprovided us with clear directions, supplementary help and zealous support that hasmade our job as painless as possible. On this note, Marek Wartenberg has fulfilled hisrole as our advisor by providing guidance and assistance during our projectdevelopment.Secondly, Bruce R. Land of Cornell University has posted an invaluable source onthe construction of analog neurons that has given us a starting point for everything.From here, we were able to gather the basics of analog neurons and gather enough datato construct our own neural analogs.Dr. Land provided us with basic neuralcomponents that have been further modified to construct our own specific alternatives.Dr. Land’s work saved us from a lot of theoretical and research-based work, allowing usto focus on more design-oriented aspects.Lastly, James M. Bower and David Beeman, authors of The Book of GENESIS,supplied us with some theoretical physiology and modeling techniques which haveproved to be absolutely essential to the construction of our analog system. The Book ofGENESIS, in and of itself, has provided our team with a “gold standard” to ensure thatour developing system as close as possible to this already accepted theory anddescription of natural phenomena. Without this work, we would have little to base ourprogression off.

40 BME 4900 Final ReportTeam 812 Appendix12.1 Project SpecificationsMechanical:Size:Electrical:Maximum Input Current:Maximum Output Voltage:Not specific; small enough to move by hand.150 microamps (A) (scalable if necessary)100 millivolts (mV) (scalable if necessary)Environmental:Storage Temperature: 60 - 90 °FOperating Temperature: 60 - 90 °FOperating Environment: Indoors (Laboratory, Clinical)Software (for data acquisition):User interface:Hardware Interfaces:Computer Requirements:Operating System:Processor:Memory (RAM):Safety:Maintenance:Oscilloscope, Keyboard, Mouse, LabVIEWOscilloscope, Monitor, NI DAQ inputsWindows 7/Vista/XP SP2, Mac OS X 10.5 or laterPentium 4/M or better (Windows)Intel-based processor (Mac OS X)1 GBDamage to the device or user may occur if inputsare not properly connected to the system. Primarydangers include electrocution, destroying circuitcomponents, and minor burns. No special safetyequipment should be required.The circuitry should be kept clean, particularly ofdust or residues forming on circuit elements orcontacts for inputs, wires, or nodes.

12.2 Purchase Requisitions and Price QuotesBME 4900 FINAL REPORT 41Team 8

42 BME 4900 Final ReportTeam 812.3 MiscellaneousAt this time, no additional information is needed in this section.