ERP Measurement Tool for MNE

[jdkoen]

The Evoked object only has methods to return peak amplitude and latency at present. Is there any interest in the development of additional amplitude and latency measures (e.g., average amplitude in a time window, fractional peak latency)? My thought would be to port the some of the measures implemented in ERPLAB's ERP Measurement Tool to MNE to help facilitate ERP analysis using MNE.

I am happy to lead development of this. These features could be added as methods to the Evoked object, or alternatively be included as a separate library. Is there a preference on which route is best?

Also, while I have coded in Python before, this will be my first time contributing to any Python software development. Thus, I would likely need some help and guidance on navigating development of unit tests and other aspects of contributing.

[agramfort]

hi @jdkoen my suggestion would be to start small with what you consider the most key feature of ERPLAB. What would that be? We'll be happy to support you.

[jdkoen]

Thanks @agramfort! One of the key features (at least to me) is pulling out mean amplitude in a user defined time window. I will start with that, and post here if I have questions or need help.

[agramfort]

I would do this with a .mean() method on Evoked like we have a .mean() method on stc

…

[jdkoen]

Thank you for the suggestion. I just noticed that the .get_peak() method returns the peak amplitude/time over an entire class of channels (i.e., the peak across all channels). Would it be useful to have a different function that extracts the peak amplitude/voltage for a user-specified set of channels? I was unsure of the rationale for taking a 'global' peak value/latency. The reason I ask the above question, and more generally, is I wonder if it will be useful to create a new class to deal with extracting data from evoked objects. From what I have in mind, it would require ~5-6 new methods on the Evoked class (with associated private functions). Currently familiarizing myself with the code structure, but wanted to ask this now before I start to code.

…

On Fri, May 29, 2020 at 3:19 AM Alexandre Gramfort ***@***.***> wrote: I would do this with a .mean() method on Evoked like we have a .mean() method on stc > — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#7848 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAXXPVF5ZEBID2KUPEH6NPLRT5OZNANCNFSM4NNL24NQ> .

[drammock]

@jdkoen for a peak over a subset of channels, there are methods evoked.pick, evoked.pick_channels, and evoked.pick_types; these return an evoked instance with a reduced number of channels, so they can be chained with get_peak like so: evoked.copy().pick_channels(['Cz', 'Fz', 'C3', 'C4']).get_peak()

Note the use of copy() because otherwise the various pick methods will modify evoked in place.

[jdkoen]

Thanks @drammock for that feedback. That approach just slipped my mind. Makes a lot more sense now for how to format the mean signal measure I'm trying to add.

[mmagnuski]

One of the key features (at least to me) is pulling out mean amplitude in a user defined time window.

Chaining can be useful here too:

evoked.copy().crop(tmin=0.09, tmax=0.11).pick_channels(channel_list).data.mean()

will give you average over specified time window and channel_list.

Maybe a good first step would be to add a short example/tutorial dealing with just a few of these measures and then we'll see if some turn out to be too complex to obtain with what is available in mne now.

[larsoner]

Maybe a good first step would be to add a short example/tutorial dealing with just a few of these measures and then we'll see if some turn out to be too complex to obtain with what is available in mne now.

I think we should make a tutorial on ERP-style analyses that contain 90% of the analyses people would be interested in. Even if it's long, a .. contents:: at the top will make it easy for people to find what they're looking for. This exercise would make it easy to see what we should turn into a separate function in MNE, if anything.

[larsoner]

@sappelhoff or @mmagnuski up for coding this in the next couple of weeks by any chance?

[sappelhoff]

I think we should make a tutorial on ERP-style analyses that contain 90% of the analyses people would be interested in. Even if it's long, a .. contents:: at the top will make it easy for people to find what they're looking for.

I am pretty packed so would be grateful if somebody else can take the time. But I agree that this would be very useful to have.

[mmagnuski]

@larsoner Sorry, I won't be able to do this for the next release. But I will have more time in about 1.5 months - then I could help with such tutorial. :)

[larsoner]

Sounds good. Or @jdkoen if you've made some progress, you could also take a stab at it if you want!

[jdkoen]

Thanks. I’m happy to take a go at it. Sorry for the delay getting back to this.

…

On Tue, Aug 25, 2020 at 7:23 PM Eric Larson ***@***.***> wrote: Sounds good. Or @jdkoen <https://github.com/jdkoen> if you've made some progress, you could also take a stab at it if you want! — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#7848 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAXXPVE4BYDFCSVXVP2JHILSCRBWNANCNFSM4NNL24NQ> .

[jdkoen]

Here are my thoughts on what measures to add for a tutorial: 1) Peak amplitude/latency for a single channel 2) Mean amplitude in a time interval 3) Fractional Area Latency (for an example, 50% area) 4) Fractional Peak Latency (to measure onsets) There are some other measures (e.g., integral measures) that I am not looking at here. Also, for peak measures (1, 3, and 4 above) there should likely be options to find absolute, positive, and negative peaks. Any thoughts on this?

…

On Tue, Aug 25, 2020 at 7:25 PM Joshua Koen ***@***.***> wrote: Thanks. I’m happy to take a go at it. Sorry for the delay getting back to this. On Tue, Aug 25, 2020 at 7:23 PM Eric Larson ***@***.***> wrote: > > > Sounds good. Or @jdkoen <https://github.com/jdkoen> if you've made some > progress, you could also take a stab at it if you want! > > > > > — > You are receiving this because you were mentioned. > Reply to this email directly, view it on GitHub > <#7848 (comment)>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/AAXXPVE4BYDFCSVXVP2JHILSCRBWNANCNFSM4NNL24NQ> > . > > >