FFAI Brain Machine Interfaces

Get Started. It's Free
or sign up with your email address
Rocket clouds
FFAI Brain Machine Interfaces by Mind Map: FFAI Brain Machine Interfaces

1. object decoding

1.1. Meyers et al. Science

1.2. stimulus

1.3. task: cat/dog decision

1.4. multi-unit recording from 443 ITC and 525 PFC were recorded

1.4.1. Inferior Temrporal Cortex

1.4.2. Prefrontal Cortex

1.5. decoding approach

1.5.1. train a classifier on the neural output activity

1.5.2. predict the class from that output activity

1.5.3. nearest neighbor classifier based on correlation coefficient

1.6. questions

1.6.1. what information is present in a particular area, and when is it present?

1.6.2. how is that information represented?

1.7. results

1.7.1. sample-stimulus identity

1.7.2. sample stimulus category

1.7.3. decision stimulus category

1.7.4. match-nonmatch

1.7.5. comparison

1.7.5.1. labels

1.7.6. coding, ITC

1.7.7. coding, PFC

2. cyborgs

2.1. basic questions

2.1.1. can we rejuvenate by transplanting brains?

2.1.2. can we combine human brains and robot bodies?

2.2. not new questions

2.2.1. head transplant - feasible today, result would be quadriplegia; rejection drugs allow this, BCI allows better quality of life

2.2.2. brain transplant - theoretically possible; fewer rejection problems, BCI essential; classical "brain in a box"

2.2.3. partial brain transplant - easier or harder, requires stem cells

2.3. physical immortality?

2.3.1. brain ages and deteriorates

2.3.2. it's unknown whether this is intrinsic to the brain or caused primarily by the body

2.3.3. it's unknown how much the brain is dependent on a functioning body (e.g., for stem cells)

2.4. brain transplant (Robert White, Cleveland Medical Hospital)

3. neuroscience and neurosurgery

3.1. gross structure

3.2. individual neurons

3.3. awake brain surgery

3.4. main points

3.4.1. the brain is divided into areas with highly specific and conserved functions

3.4.2. the brain itself is composed of neurons; each neuron operates both chemically and electrically, and these electrical activities can be recorded and stimulated

3.4.3. electrical activity in the motor cortex generates signals that are conducted via nerves to muscles

3.4.4. light arriving in the eye generates nerve signals that travel to the visual cortex and are processed there

3.5. general points

3.5.1. brain and nervous system anatomy is extremely complicated and sensitive

3.5.2. things that sound nice theoretically ("let's interface directly with the brain") are messy, risky, and complicated

3.5.3. there is a big need for brain computer interfaes for the disabled

3.5.4. brain surgery has made enormous advances

4. why?

4.1. UIs

4.2. helping the disabled

4.3. helping locked-in patients

4.4. cyborgs / brain in a box

4.5. neuroscience research

4.6. accessing brain states directly

4.7. application of machine learning

5. optogenetics

5.1. optogenetic stimulation Nature 2010 Video

5.2. optogenetic recording

6. direct brain interfaces

6.1. multielectrode arrays

6.1.1. brainstem implant

6.1.2. Utah electrode array

6.1.3. implantation sites

6.1.4. problems

6.1.4.1. inflammation due to friction and stiffness (insertion requires stiffness)

6.1.4.2. inflammation due to unnatural materials

6.1.4.3. limited density

6.1.4.4. limited spatial resolution, fixed geometry

6.1.4.5. multiple neural targets (more a problem for stimulation than recording)

6.1.4.6. transcutaneous connections, power, etc.

6.1.5. solutions

6.1.5.1. biocompatible surface coatings

6.1.5.2. stiffness during insertion with biodegradable matrix

6.2. output

6.2.1. ECOG arm

6.2.1.1. ECOG background

6.2.2. process

6.2.2.1. electrodes are hooked up to motor areas (the same areas that control muscles)

6.2.2.2. human learns to generate the necessary control signals similar to learning a regular motor skill

6.2.3. status

6.2.3.1. experimental, costly, stationary

6.2.3.2. functional, potentially useful

6.2.3.3. other potential application: electronic bridging of severed spinal cord

6.3. input

6.3.1. auditory

6.3.1.1. cochlear implant

6.3.1.2. brainstem implant

6.3.1.3. process

6.3.1.3.1. fairly easy mapping of sounds into electrical signals

6.3.1.4. status

6.3.1.4.1. routine

6.3.1.4.2. high quality perception possible

6.3.2. visual

6.3.2.1. retinal implants

6.3.2.2. cortical implant

6.3.2.3. process

6.3.2.3.1. difficult, manual mapping of cortical locations to spatial locations

6.3.2.3.2. (how does this happen naturally?)

6.3.2.4. status

6.3.2.4.1. highly experimental

6.3.2.4.2. poor quality, no real perception, but still useful

7. fMRI

7.1. fMRI background

7.2. fMRI (and neuromarketing)

7.3. lie detection

7.4. fMRI BCI

7.5. fMRI reconstruction from visual cortex

8. EEG

8.1. EEG headband

8.1.1. Interaxon Muse

8.1.2. Neurosky

8.1.3. BCI typing

8.1.4. Emotiv

8.2. EEG

8.2.1. example

8.2.2. measures correlated synaptic activity of post-synaptic potentials in the cortex

8.2.3. high temporal resolution

8.3. EEG signals

8.3.1. different "waves"

8.3.2. delta waves - slow-wave sleep, attention (4 Hz)

8.3.3. theta waves - idle (4-8 Hz)

8.3.4. alpha waves - relaxation (8-13 Hz)

8.3.5. beta waves - alertness (13-30 Hz)

8.3.6. gamma waves - complex perceptual tasks (30 Hz and higher)

8.3.7. irregular - transient effects, states; epilepsy

8.3.8. spatial distribution

8.4. algorithms

8.4.1. generally, collection of signals from many channels

8.4.2. each channel is a different mixture of multiple sources

8.4.3. problem: "blind source separation"

8.4.4. solution: ICA (independent component analysis), assuming statistical independence of the different components

8.5. Necomimi

8.6. applications

8.6.1. biofeedback, relaxation

8.6.2. pure brain-based interfaces

8.6.3. lie detection