As a lot of interest has continued to develop in NeuroCARE Pro (or NCP), I thought it might be useful and interesting to start a thread here about it as well. A good starting point for some information about NCP is our webiste:

www.zengar.com

and the zengarncp@yahoogroups.com forum.

NCP uses comprehensive, adaptive targeting that is based on real-time Gabor transforms. Gabor transforms are the fundamental "mechanism" underlying human perception (as described by Pribram quite a while go now!) and are elegant mathematical tools for extracting/recognizing the information value of a process. Information about Gabor transforms -- and Joint Time-Frequency Analysis more generally -- can be found in a variety of places, including:

http://zone.ni.com/devzone/conceptd.nsf/webmain/BD7D314EE0E3B87086256866005A7D66?OpenDocument

or just a Google search on EEG Gabor and the like will turn up some interesting possibilities.

It's an elegant way to work.

val

One question that freqeuntly arises in re: to NCP is whether or not "interhemispheric" training can be done using NCP. Yes it can -- and with an enormous degree of variety and modification.

Interhemispheric training has been supported in NCP since the beginning; however, most users find that, once they use our default ways of targeting, they realy don't find as much use for interhemispheric work per se. One way to implement interhemispheric training is to setup each channel's sensor array as a bipolar montage -- say T3 (active) T4 (reference) A1 (ground). This is a very simple way to begin, esp if you are not using v1.9.

If you are using v1.9 you can simply toggle on "Difference" based targeting and this will specifically compute the difference between each of the targets in the bilateral array of target being used.

One can even tweak the filtering opeations "on the fly" to emulate the kind of elliptic IIR filters that are found on other systems. And if you want to "nudge" the center point of a particular target, you can use 0.1 Hz increments for the nudge. Or totally remap the frequency range and other filter parameters WHILE actually training your client.

There's a lot to NCP that many people simply don't know.

val

Val writes:
Interhemispheric training has been supported in NCP since the beginning; however, most users find that, once they use our default ways of targeting, they really don't find as much use for interhemispheric work per se.

Siegfried answers:
I suspect that NCP does not at this moment actually implement inter-hemispheric training as we know it, but I would love to be wrong about this.

Val writes:
One way to implement interhemispheric training is to setup each channel's sensor array as a bipolar montage -- say T3 (active) T4 (reference) A1 (ground). This is a very simple way to begin, esp if you are not using v1.9. [Let's call this Mode A"”Siegfried]

Siegfried answers:
This certainly sets up the signals we are interested in, but in my understanding NCP works mainly with threshold crossings: e.g., the AVI goes or doesn't go. The signal dynamics are largely lost in this process. Now I am very impressed and surprised with what can be done with inhibit-based, threshold-driven training, but it does not focus on the high dynamics that we like to present to the person (through the variable brightness of Pacman, the size of Boxes, etc.).

Val writes:
If you are using v1.9 you can simply toggle on "Difference" based targeting and this will specifically compute the difference between each of the targets in the bilateral array of target being used. [Let's call this Mode B"”Siegfried]

Siegfried answers:
I am tempted to conclude that the working signal here is the output of the Gabor transform for the two channels. That would be roughly the equivalent of looking at the envelope function of the reward waveform (in NeuroCybernetics you can get to that on the clinician display with "alt F10", as I recall, and with F12 [which is unsupported in the old NC, but it usually works]). In any event, if that assumption is correct, then the phase of the signal is not being processed, and therefore does not enter the calculation of the relationship between the two channels. That leaves out the very thing that gives inter-hemispheric training its value, and on that basis I am not at all surprised that people who try this on NCP won't be particularly impressed.

Val writes:
One can even tweak the filtering operations "on the fly" to emulate the kind of elliptic IIR filters that are found on other systems. And if you want to "nudge" the center point of a particular target, you can use 0.1 Hz increments for the nudge. Or totally remap the frequency range and other filter parameters WHILE actually training your client.

Siegfried answers:
If the output of the filtering operation using the IIR filters can be made available to the difference targeting mode above [Mode B]"”meaning that the calculation has to be done at high rep rates to cover the frequencies of interest"”and if the resulting signal can be made available to the trainee with its prevailing dynamics on the client screen (some smoothing will be necessary, of course), or via auditory/tactile feedback, then I think the value of NCP will be considerably enhanced for all of those who do our type of training.

Alternatively, if the elliptic IIR filters could be used in Mode A above, then the resulting feedback signal will incorporate the phase information of interest, and then it is simply a matter of representing that dynamic signal to the client. Perhaps auditory or tactile feedback would be best for this, in order not to impinge upon the present design of visual presentation.

Siegfried

Sieg:

Probably each of these points should be a separate post. And you're actually incorrect in most of the points but, as I've said on many occasions before, NCP is really not well known by most...

val

Val writes:
One way to implement interhemispheric training is to setup each channel's sensor array as a bipolar montage -- say T3 (active) T4 (reference) A1 (ground). This is a very simple way to begin, esp if you are not using v1.9. [Let's call this Mode A"”Siegfried]

Siegfried answers:
This certainly sets up the signals we are interested in, but in my understanding NCP works mainly with threshold crossings: e.g., the AVI goes or doesn't go. The signal dynamics are largely lost in this process. Now I am very impressed and surprised with what can be done with inhibit-based, threshold-driven training, but it does not focus on the high dynamics that we like to present to the person (through the variable brightness of Pacman, the size of Boxes, etc.).

Actually you're wrong on this. We have pre-packaged Portals that specifically do use proportional feedback -- both with single threshold lines, as well as with boxes. And, by default, those targets are autoadjusting but that can be turned off while training so that more traditional "static" proportional feedback can actually be done. It's just that most users find they don't want to actually do that once they being using our more typical ways of targeting.

We also use free-form visualizations that do demonstrate visually the dynamics of the signals of interest. So, I know that you like to think of what we do as "inhibit" based on "on-off" but we don't even have "inhibts" per se in the system (except in the sense that the computer itself can be inhibited from producing sounds et al).

No, NCP is a much "deeper" box than you're used to thinking it is.

Val writes:
If you are using v1.9 you can simply toggle on "Difference" based targeting and this will specifically compute the difference between each of the targets in the bilateral array of target being used. [Let's call this Mode B"”Siegfried]

Siegfried answers:
I am tempted to conclude that the working signal here is the output of the Gabor transform for the two channels. That would be roughly the equivalent of looking at the envelope function of the reward waveform (in NeuroCybernetics you can get to that on the clinician display with "alt F10", as I recall, and with F12 [which is unsupported in the old NC, but it usually works]). In any event, if that assumption is correct, then the phase of the signal is not being processed, and therefore does not enter the calculation of the relationship between the two channels. That leaves out the very thing that gives inter-hemispheric training its value, and on that basis I am not at all surprised that people who try this on NCP won't be particularly impressed.

val: This is precisley wrong as the Adaptive Gabor is precisely phase perserving in all operations; moreover, our preferred method for calculating the "Current Value Bar" is fourth moment about the mean (when using Gabor-based targeting). This means we are specifically looking at and responding to the "shape" of the phase changes as they are occuring -- either in difference mode or in each separate target, in a parallel mode of calculation.

Val writes:
One can even tweak the filtering operations "on the fly" to emulate the kind of elliptic IIR filters that are found on other systems. And if you want to "nudge" the center point of a particular target, you can use 0.1 Hz increments for the nudge. Or totally remap the frequency range and other filter parameters WHILE actually training your client.

Siegfried answers:
If the output of the filtering operation using the IIR filters can be made available to the difference targeting mode above [Mode B]"”meaning that the calculation has to be done at high rep rates to cover the frequencies of interest"”and if the resulting signal can be made available to the trainee with its prevailing dynamics on the client screen (some smoothing will be necessary, of course), or via auditory/tactile feedback, then I think the value of NCP will be considerably enhanced for all of those who do our type of training.

val: This is -- and has been -- the case for those who were interested in doing so....

However, I would again point out that of those who have used both systems (and both in their own "native" modes), almost all come back to NCP.

Alternatively, if the elliptic IIR filters could be used in Mode A above, then the resulting feedback signal will incorporate the phase information of interest, and then it is simply a matter of representing that dynamic signal to the client. Perhaps auditory or tactile feedback would be best for this, in order not to impinge upon the present design of visual presentation.


This also is -- and has been -- included in NCP for quite some time.

Well then, with all of these capabilities in NeuroCarePro, it should be a simple matter to dial up one of the relevant portals and replicate what we are doing with Interhemispheric Training.

Perhaps you could demonstrate that for everyone's benefit at the Winter Brain Conference.

I would love to see a combination of the usual 16-target strategy with a display on the client screen of the differential (bipolar) signal, derived from C3-C4 or T3-T4, or even (C3-A1) minus (C4-A2), with high signal dynamics.

The dynamic reward signal could be shown as a box of variable size like in the NeuroCybernetics Boxes game, or it could be shown as a circle or disc.

For the equivalent for two channels, you could show an ellipse, as in the BrainMaster two-channel display, with each channel controlling one of the major axes of the ellipse.

Or the signal could be delivered to headphones, or made available at the soundcard to drive a vibro-tactile feedback implementation.

While you are at it, perhaps you could port the Lissajous two-channel display to the client screen as well, so that trainees can see the relative phase relationships of the two-channel signals directly.

Specifically, the reward signal should be derived from elliptic IIR filters with two poles, with selectable center frequency and selectable width.

The combination of the 16-target strategy with a dynamic reward should be quite attractive to a large number of practitioners.

Siegfried

Have done and can do again except not in addition to the 16 targets; however, that actually IS doable as well.

I'll have to dig out the "old" displays for that kind of stuff. What we find is that once people use NCP they don't go back so those approximations were of much greater interest two years ago than they are now.

there are many great functions in this software, but the stability, licensing of it are the worst I have ever seen in ANY software packages.

Further, upgrades are forced upon the user which each include their own bugs.

The research is good, the ideas and all that, but the software itself is unstable to the point where desktop machines are no longer supported.

The software is buggy and the support is done by two people who usually take around 4 days to reply to an email for licensing a new machine.

great ideas, but the product needs a complete rehaul by a professional programmer.

We are increasingly trying to replicate some of the key good ideas that are incorporated in NeuroCarePro with more accessible software, so that people have more ways to achieve comparable results. The basic ideas relate to inhibit-based training; adaptive thresholding to implement variability-based targeting; multiple channels; two-channel training; etc.

Of course we also can't quite let go of our own ideas, so we admix some reward-based training to permit fine-tuning of the training. We are at this moment close to the release of a design for the BioExplorer software that implements these ideas.

We make no claim to be replicating NCP more generally, because indeed Val has implemented lots of ideas there that he has not shared openly. And we are still partial to our own notions of how training should proceed. However, this design may well capture some of the essence of that approach for people to get acquainted with.