Neuro-Prostheses

Overview
neuroprosthetic infographic.jpeg
fig.1 Applications of Neuro-Prostheses

The field of neurally controlled prosthesis is a quickly growing and developing area of study that has the potential to transform the lives of amputees, paralyzed, and those who are blind and deaf. Through prosthetic arms, legs, cochleas, and even bionic eyes, many people who have lost their independence to move and interact with the world, have an opportunity to regain that ability. There are many types of ways that neuronal control of limbs can be applied. In paralyzed patients, brain computer interfaces can bypass the damaged area, and stimulate muscle contraction distally. There is also potential for this technology to be applied for paralyzed persons, by means of robotic limbs that are disconnected to the body or by controlling their wheelchair or computer, through neural impulses. The cochlear implants are a device that translates sound vibrations to, computer code, then to neural impulses, in order for the hearing impaired to be able to process sound. Finally, this technology can be used to use neurally controlled prosthesis. These are used primarily for amputees, and through electrode implantation on the motor cortices in the brain, are able to manipulate these bionic limbs with a lot of freedom and ease with practice. Previous technology, that control these “smart” prosthesis, have used impulses from proximal or indirect muscle activity to control the prosthetic, but by using neural impulses instead, the individual is able to achieve more intuitive and natural control over the prosthetic device [1].

Anatomy and Physiology of Movement

Motor Corticies

The motor cortex is composed of a few different areas. The primary motor cortex the premotor cortex, and the supplementary motor area. The motor cortex receives information from other cortical areas, directly and indirectly through the thalamus, and it also gets input from the basil ganglia and the cerebellum. The basil ganglia, is primarily responsible for both initiating or inhibiting movement and it sends afferent information through the thalamus to the motor cortex [3]. The cerebellum is important in calculating error and modulating movement to be more successful. The cerebellum has receptive information from proprioceptors and vestibular nuclei, that give it information on how the movement is being performed, and it has output power to modify motor commands, as they come down the descending pathways through the brainstem[3]. This makes the cerebellum vital for motor learning because of its ability to change through trial and error. It is also important in producing

homunculus.jpeg
fig. 2 Motor and Somatosentory Cortex Homunculus

fluid movements and coordination of synergistic muscle groups.
The primary motor cortex (M1), is vital to conscious control of particular or skilled movements. Located on the pre-central gyrus of the frontal lobe, the primary motor cortex has a somatotopic representation of the body on each hemisphere, called the homunculus (see fig. 2) Each hemisphere, left and right, innervates the contralateral part of the body. These cortical neurons are organized in such a way that they code for specific synergistic movements of a part of the body, rather than innervating a specific muscle [2]. Alpha motor neurons in the spinal cord enact the force of contraction of groups of muscle, whereas information from the primary motor cortex encodes parameters for movements or simple movement sequences, and deals primarily w
ith the force and direction of the overall movement[3]. The primary motor cortex also gives commands for the distance and speed of the movement [3].
The premotor cortex (PM) is located anteriorly to the pre-central gyrus in the frontal lobe and controls learned motor skills and coordinates complex task-related movements. Most of its axons are sent to the primary motor cortex where the movement is sent to be executed, about 15% of the axons, though, are sent to the descending corticospinal pathway to influence movement more directly[2]. It is also highly involved in planning movements, and uses input from sensory feedback to plan coordinated complex movement. It is important in the selection of a suitable motor plan for the voluntary m
corticospinal tract.jpeg
fig.3 Lateral and Anterior Corticospinal Pathway
ovement. Mirror neurons in the premotor cortex fire when the movement is even just visualized, or if the person is looking at a movement be performed by someone else.
The context of the movement is also important in planning, and the premotor cortex influences those actions[3].
The supplementary motor area (SMA) seems to select remembered motor sequences, and coordinates bilateral motions. Movements that are generally rehearsed and practiced and involve both sides of the body, such as playing the piano, or tying one’s shoes, require SMA involvement[3].

Descending Pathways

The descending pathways (see fig. 3) carry information from the motor cortex to the spinal cord and then to the motor neurons to the effector muscles. The main pathway is are the corticospinal pathways; anterior and lateral. These pathways originate in the premotor cortex, the supplementary motor cortex, the primary motor cortex and the somatosensory cortex. This pathway sends information about skilled and controlled movements. In a healthy person, the pathway would go from the motor cortex, through the corona radiata, to the posterior limb of the internal capsule, to the base of the cerebral peduncle, then to the corticomedullary junction (medullary pyramids). This is where the tract either decussates to the lateral corticospinal pathway or stays ipsilateral and goes on the anterior corticospinal tract. The lateral corticospinal pathway receives about 90% of the descending neurons and they travel down the spinal cord until they synapse with interneurons or the alpha motor neurons. This tract mainly controls distal musculature. The anterior corticospinal tract remains ipsilateral until the level of innervation, where it crosses the anterior white commissure and synapses on alpha motor neurons. The anterior pathway generally innervates proximal and trunk musculature[9]. If there is damage to this tract or the spinal cord in general patients lose the control of voluntary movement at and below the damaged area, and the use of neurally controlled prostheses is an option for locomotion or performance of daily tasks depending on where the injury is.

Brain Computer Interface

Brain computer interfaces (BCI) are also known as brain machine interfaces (BMI), and the ability for humans to control prostheses or cursors on a screen through them is because of the lifelong plasticity of the brain. The brain’s ability to learn and adapt due to trauma, or new devices in the environment, enables the patient to manipulate the machine. Patients have seemed to consolidate and stabilize a “prosthetic motor map” and through learning, these neurons change to be optimally compatible with the BCI and performance is increased[7]. These systems that utilize the cortex to control prostheses and require no residual motor skills to use[4].
BCI how they work.jpeg
fig.4 microelectrode implant

There are a few different ways that BCIs can receive information from your cortex. These different approaches to intercept signals from the motor cortex to control a specific prosthetic are electroencephalogram (EEG), magnetoencephalographic (MEG), mircoelectrode arrays, electrocorticography (ECoG), or epidural electrocorticography(EECoG) all have varying degrees of invasiveness, effectiveness and practicality. EEGs are noninvasive, and have electrodes placed on the skin of the skull. The drawbacks to this technique is that they are getting an indirect and weaker
electrocorticography.png
fig.5 electrocotricography electrodes on motor and sensory cortex
signal from the primary motor cortex, this gives them poor spatial resolution and less detailed information[6]. MEG signals are non-invasive, the device is placed on their head, and collect magnetic field information from dipoles created through the ions released at synaptic junctions [5].
Microelectrode arrays have tiny tips that are implanted a few millimeters into the motor cortex. This gives the user much better and more accurate control, but scar tissue has formed eliminating the electrodes effectiveness to receive information[6]. ECoG has been more successful, because it is made up of a sheet of disc electrodes that are placed directly on the motor cortex. EECoG signals come from a 32 channel electrode array that is made on a sheet thinner than plastic wrap, and it lays directly on the dura matter of the motor cortex, making it less likely to develop scar tissue. Researchers have discovered that electrodes could be millimeters apart and still have independence from one another and be controlled [6].

Neural Decoding

The next step is to translate the neural impulses read by the electrodes to precise movements in the prosthetic, and each modality requires an algorithm in the BCI to decode the intention of the movement and produce the desired result. Neural decoding algorithms are made to create a state estimate, and either create an exact classification (movement or no movement) or a prediction for continuous variables (speed, direction, and endpoint of limb)[1]. There are a few modalities to decode, and it is still up for debate which ones work best for a given medium. Artificial neural networks (ANN) are mathematical models that are made up of simulated neuronal units and axons. Each neuron receives stimulation at a certain threshold and gives a certain output value, that activates an effector. These activation levels can be linear, nonlinear, or constant, and there are functions that calculate the velocities in different axes [1]. There are also other kinds of decoders that use linear regression models to determine intended movements and these different decoders require the patient to manipulate them differently in order to use them efficiently[1,7]. Researchers have found that over several days of using different decoders, patients showed skilled performance with each decoder and even created distinct motor maps for each decoder[7].

neuroprosthpicture.jpeg
fig. 6 diagram of how BCI works to control the prosthetic

There has also been a push for decoding algorithms that learn with the patient, in order for better performance, and for the prosthetic to be able to anticipate the intended movement. This process is called closed-loop decoder adaptation (CLDA) and the program shifts and changes it’s mapping while the patient is using it to better represent the intended movement. This decoder would extract error signals from the cerebellum and change parameters of the movement accordingly. The potential problem with this method is that the brain would be adapting along with the interface, so that presents the problem of hitting a “moving target”[8].

Summary

The improvement and development of neurally controlled prostheses is an exciting field that is faced with challenges and problems, and is forced to creatively solve them. Through this process researchers have been able to understand the way that we move and operate much better, but there is still a lot that we don’t know. It is very hard to have a machine controlled by a brain perform natural moments, and because development of proprioception and sensory receptors is still underway, patients are might only be receiving visual feedback about their movements. There is development of prostheses that can "feel" and send afferent sensory data to the cortex, which will probably be a reality in the near future.
brain-computer-interface-in-action.jpg
fig. 7 Woman controls robotic arm through BCI


Glossary

Actuator: a prosthetic device that is not connected to the body
BCI or BMI: (Brain Computer Interface or Brain Machine Interface) the relationship between autonomous neural impulses and a computer’s output.
Closed Loop: the patient receives feedback, either visual or sensory feedback about the movement or position of the prosthetic
EEG: (electroencephalogram) noninvasive, and have electrodes placed on the skin of the skull
EECoG: (epidural electrocorticography) a 32 channel electrode array that is made on a sheet thinner than plastic wrap, and it lays directly
on the dura matter of the motor cortex, making it less likely to develop scar tissue.
ECoG: (electrocorticography) a sheet of disc electrodes that are placed directly on the motor cortex.
MEG: (Magnetoencephalographic) these signals, collected noninvasivley and in real time, from dipoles created through ions released at synaptic junctions in the brian and connect to the BMI and control a prosthetic device
Neural Decoder: an algorithm or regression that translates the neural impulses to appropriate electrical signals
Neuro-plasticity: the brain’s ability to adapt, reorganize and learn in response to challenges or changes in output or sensory input
Open-Loop: the patient receives no feedback about the movement or position of the prosthetic during testing
Prosthetic: an abiotic tool that is used to replace or enables a non-functioning or missing part of the organism

Quiz

True False:

  1. T/F The neural connections in the motor cortex are static and unable to change once you reach adulthood
  2. T/F Patients can produce a prosthetic motor map through practice and learning to use a BCI
  3. T/F EECoGs are placed onto the dura mater of the cortex

Multiple Choice:

  1. Which of the following are noninvasive (no surgical opening of the skull) methods of collecting neural impulses.
    • a. EEG
    • b. EECoG
    • c. Microelectrodes
    • d. ECoG
  2. Which of the following do not send information to the motor cortex?
    • a. SMA
    • b. Premotor cortex
    • c. Somatosensory Cortex
    • d. Basil Ganglia
    • e. All of the above send info to the cortex
  3. What which one of these BCI methods works well at first, but produces scarring and is ineffective after time?
    • a. ECoG
    • b. EEG
    • c. microelectrodes
    • d. MEG

Short Answer

  1. How do most patients receive feedback about location of limb and success of movement with neurally controlled prostheses now? How could that be improved?
  2. What is the purpose of the neural decoder?


Further Readings

http://www.the-scientist.com/?articles.view/articleNo/41324/title/Neuroprosthetics/
An article about the basics of the Neuroprosthetics field and some developments.

https://www.youtube.com/watch?v=ZuATvhlcUU4
TEDx Talk on Neural Prosthetics and how they work (gives a lot of good visuals, and is explained clearly)

https://www.youtube.com/watch?v=QRt8QCx3BCo
Video that shows paralyzed patient use BCI to control robotic arm

Sources

  1. White JR, Levy T, Bishop W, Beaty JD (2010) Real-Time Decision Fusion for Multimodal Neural Prosthetic Devices. PLoS ONE 5(3): e9493. doi:10.1371/journal.pone.0009493
  2. Marieb, E., & Hoehn, K. (2011). The Central Nervous System. In Anatomy & physiology (4th ed.). Upper Saddle River, NJ: Pearson/Benjamin Cummings.
  3. Knierim, J. Motor Cortex: Neuroscience Online. Retrieved December 13, 2015, from http://neuroscience.uth.tmc.edu/s3/chapter03.html
  4. Chirimuuta, M. (2013). Extending, Changing, and Explaining the Brain. Biology and Physiology, 28(4), 613-638. doi:10.1007/s10539-013-9366-2
  5. Fukuma R, Yanagisawa T, Yorifuji S, Kato R, Yokoi H, Hirata M, et al. (2015) Closed-Loop Control of a Neuroprosthetic Hand by Magnetoencephalographic Signals. PLoS ONE 10(7):e0131547. doi:10.1371/journal.pone.0131547
  6. Washington University in St. Louis. (2011, February 19). Mind over matter: EECoG may finally allow enduring control of a prosthetic or a paralyzed arm by thought alone. ScienceDaily. Retrieved December 14, 2015 from www.sciencedaily.com/releases/2011/02/110218142440.htm
  7. Ganguly K, Carmena JM (2009) Emergence of a Stable Cortical Map for Neuroprosthetic Control. PLoS Biol 7(7): e1000153. doi:10.1371/ journal.pbio.1000153
  8. Carmena JM (2013) Advances in Neuroprosthetic Learning and Control. PLoS Biol 11(5): e1001561. doi:10.1371/journal.pbio.1001561
  9. Knierim, J. Spinal Reflexes and Descending Motor Pathways: Neuroscience Online. Retrieved December 16, 2015, from http://neuroscience.uth.tmc.edu/s3/chapter02.html

Image Sources

  1. http://cint.wustl.edu/graphics/assets/images/Projects/image353.jpg
  2. https://c1.staticflickr.com/9/8402/8641421501_c07dd92da4_b.jpg
  3. http://what-when-how.com/wp-content/uploads/2012/04/tmp14104_thumb2.jpg
  4. http://www.techlegends.in/wp-content/uploads/2014/07/brain-computing-working-300x300.gif
  5. https://upload.wikimedia.org/wikipedia/commons/thumb/e/e5/Intracranial_electrode_grid_for_electrocorticography.png/220px-Intracranial_electrode_grid_for_electrocorticography.png
  6. http://biomed.brown.edu/Courses/BI108/BI108_2003_Groups/Spinal_Cord/images/neuroprosthpicture.jpg
  7. http://images.scienceworldreport.com/data/images/full/3068/brain-computer-interface-in-action.jpg?w=680

Quiz Answers:

True/False
  1. F
  2. T
  3. T
Multiple Choice
  1. a.
  2. e.
  3. c
Short Answer
  1. Visual feedback is the sole sort of feedback patients get, but they are developing prostheses that can feel and send sensory information to the somatosensory cortex, to aid in production of more natural movements
  2. A neural decoder translates the an algorithm or regression that translates the neural impulses to appropriate electrical signals in the BCI.