bci2000.org BBS

Hi All,
I have been reading a lot about the BCI systems and the more i read the more i get confused as to what is the difference between classifiers and translation algorithms are they one and the same thing or different please elaborate...
Maha Syed

This BBS has members that have much more expertise in classification/translation than I. If anyone else responds to this post contradicting anything I have said then you should assume that they are correct and that I am wrong.

In my understanding, a classifier is generally used to assign a brain signal segment to one of a few classes. For example, a response to a flash on the screen might be classified as either a P300 (class 1) or Non-P300 (class 0). A translation algorithm is used to translate a brain signal segment into a continuous variable used for continuous control. For example, the firing rate of a neuron might be translated into the speed of a computer cursor in the x-direction.

The distinction becomes murky when you use the output of a classifier (before it has been thresholded) as a continuous variable. For example, you might have a classifier that will determine if a particular segment of EEG belongs to a "motor-imagery" class (class 1) or a "not motor-imagery" class (class 0). Usually this is done by first extracting a feature from the EEG. Let's say that feature's value for a particular EEG segment is 0.63. In the past you determined that setting a threshold of 0.45 provides the most accurate classification into either class 0 (x < 0.45) and class 1 (x >= 0.45). Thus this EEG segment would be considered class 1. However, you can use that value (0.63) directly to control some feedback (e.g., the speed of a computer cursor) and suddenly your classifier has become a translation algorithm.

If you want to learn more, I suggest you start with this excellent and free online course about Machine Learning: https://www.coursera.org/course/ml It doesn't deal directly with BCIs but it will really help you get a deeper understanding of classification/regression.

[Edited to fix a few typos]

Thanx i will be needing loads of help as i am working on a feasability of "Non-invasive BCI systems for visually impaired"...currently i am just trying to understand what BCI systems are and how they work and time to time loads of questions come up I also wanted to know that the BCI systems i have come accross work in one direction i.e the data (signal/ order) from the brain is taken - translated into device commands..
wanted to know is there any system in which bi-directional communication is also taking place e.g. taking pictures converting them to electrical signals and then sending these signals to brain so that pictures can be seen by the brain bypassing the eyes... is it possible yet... has research been done in this area so far?

mahasyed wrote:I also wanted to know is there any system in which bi-directional communication is also taking place e.g. taking pictures converting them to electrical signals and then sending these signals to brain so that pictures can be seen by the brain bypassing the eyes... is it possible yet... has research been done in this area so far?

Recently a researcher has stimulated monkey brains to simulate the perception of texture but I wonder if the monkey actually perceives texture or if they simply perceive a difference in the stimulation patterns. Maybe that is a distinction without a difference. See here: http://www.nature.com/news/2011/111005/ ... 1.576.html

By far the more common and successful way to get information into the brain is to use existing senses or by stimulating peripheral nerves directly. Cochlear implants are a ubiquitous example of this. Some vision prosthetics have been developed that capture images with a camera then stimulate the tongue or the retina http://www.nature.com/news/restoring-si ... ts-1.10627.

I wouldn't really consider these systems "bi-directional" because they are mostly one-way (into the nervous system). I would definitely not consider them BCIs because a user's brain signals are not being used to manipulate the user's environment.

The only true bi-directional (brain -> computer -> brain) device that I know about is the Neurochip2 http://www.ncbi.nlm.nih.gov/pubmed/21632309. It's a fantastic tool for neuroscience but I don't yet know of any medical applications.

I'll only mention one more thing because it's close to my field of study. If you use a brain signal to move a prosthetic limb then this is mostly one-way because, other than vision, the brain doesn't have much information about what the limb is doing or where it is in relative space. If you use a brain signal to move an exoskeleton attached to a paralyzed limb then you might get a little more information to the brain if the proprioceptive systems are still intact (in stroke, certain spinal cord lesions, and ALS) but the effectiveness of proprioception is diminished considerably without the corticospinal influence on gamma motoneurons. Finally, you can use a brain signal to control functional electrical stimulation of a muscle http://www.ninds.nih.gov/news_and_event ... alysis.htm but this may or may not produce proprioception depending on the stimulus parameters. If you combine the above systems with electrical stimulation of a peripheral nerve then you can enhance proprioception and now you're approaching bi-directional.

In cases where the proprioceptive pathway is also damaged (e.g., spinal cord transection), the only way to get information about the limb position to the brain would be through another modality (like vision or hearing), by stimulating another peripheral nerve, or by stimulating proprioceptive pathways in the CNS.

I think Lee Miller alludes to this in his Nature Neuropod interview. http://www.nature.com/neurosci/neuropod ... 04-27.html