# Comparison between original EEG and source activity signal

**URL:** <https://mne.discourse.group/t/comparison-between-original-eeg-and-source-activity-signal/3143>\
**Category:** Support & Discussions\
**Created:** [May 16, 2021, 3:10pm UTC](https://mne.discourse.group/t/comparison-between-original-eeg-and-source-activity-signal/3143 "2021-05-16T15:10:03Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![Davi1990](https://avatars.discourse-cdn.com/v4/letter/d/54ee81/32.png) [@Davi1990](https://mne.discourse.group/u/Davi1990)\
**Post date:** [May 16, 2021, 3:10pm UTC](https://mne.discourse.group/t/comparison-between-original-eeg-and-source-activity-signal/3143/1 "2021-05-16T15:10:03Z")

</div>

Dear MNE expert,

I’ve computed the source activity and now I want to bring it back to channels space. For this purpose, I’ve run the following code:

```python
EEG = mne.apply_forward(fwd_fixed, stcs, real_EEG_epoched.info).data 

```

However, if I plot the original EEG data (the one that I used to compute the source activity) with the data that I got from this transformation (source activity to channels space) they look different. In the following image I’ve plotted in red a channel from the original EEG and in blue the same channel obtained by this transformation

![Screen Shot 2021-05-16 at 11.04.40](https://global.discourse-cdn.com/free1/uploads/mne/original/1X/3fecd3dc5af2e724329f8167558163f0a55decd2.png)

Given the nature of the inverse problem,

> M = GX + E

I was wondering whether if it’s expected that the two signals are different. If yes, why is that?

Is there a way to retrieve the original EEG signal from its source estimation?

Thanks in advance

---

<div class="post-metadata">

**Author:** ![agramfort](https://avatars.discourse-cdn.com/v4/letter/a/7c8e57/32.png) [@agramfort](https://mne.discourse.group/u/agramfort)\
**Post date:** [May 17, 2021, 7:32am UTC](https://mne.discourse.group/t/comparison-between-original-eeg-and-source-activity-signal/3143/2 "2021-05-17T07:32:38Z")

</div>

I see 2 reasons for the differences:

- the inverse problem can use the dipoles non normal to the cortex to explain the data
- you regularized a lot in the inverse problem with a strong lambda2

Alex

---

<div class="post-metadata">

**Author:** ![Davi1990](https://avatars.discourse-cdn.com/v4/letter/d/54ee81/32.png) [@Davi1990](https://mne.discourse.group/u/Davi1990)\
**Post date:** [May 17, 2021, 12:04pm UTC](https://mne.discourse.group/t/comparison-between-original-eeg-and-source-activity-signal/3143/3 "2021-05-17T12:04:26Z")

</div>

@agramfort

lambda2 was default. I’ve seen someone has the same issue in the past:

> [@Forward/Inverse modeling sanity checks](https://mne.discourse.group/t/forward-inverse-modeling-sanity-checks/2559):
>
> Objective I’m playing around with forward/inverse modeling in MNE and am trying to check the assumption: (1) kernel @ gain == I Where kernel is the prepared inverse operator kernel, gain is the leadfield matrix, and I is an identity matrix. By extension, I want to check that: (2) x == kernel @ gain @ x Where x is the original time series data. (2) is easier to check due to how MNE stores the leadfield matrix, but I need to obtain kernel and gain that satisfy (1) as well. Assumptions f…

Given the reasons for the differences you listed, any suggestions on how to retrieve the original EEG signal from a practical point of view?

Thanks