[Eeglablist] (Updated) Huge challenge: Dipoles uncorrelated with components

[Curtis Bingham]

(UPDATE with links to images, and additional information in UPDATES section below, as requested by kind souls who responded privately)

 

Why are my dipoles completely uncorrelated with independent components? 

In this image (https://urldefense.com/v3/__https://imgur.com/wlHSs0X__;!!Mih3wA!W92KXdh7blda6IzoYtjvGjRu6xaTeySMzUMtOXyW_F1oJlZEaDjOLADtJxU5Hp4eJLIK6g$ ), you can see that the dipoles representing the components are almost entirely uncorrelated with the independent components. 
The dipole for IC 7 is the only dipole properly correlated with the underlying component
Dipoles for IC 2 and IC 9 are in opposite hemispheres from the components
The remaining dipoles are a LONG way away from their components. 
 

What would cause this extreme separation of dipole from independent component?

 

The EDF file is generated by Neurofield Q20 and processed as follows: import, channel locations (10-20 or 10-05 locs file makes no difference) average reference, remove baseline, high and low-pass filter, runica, ICLabel, and delete rejected components. 

 

DIPFIT is run with Template BEM (MNI), no co-reg, autofit (RV%=100)

There is a warning in each pass saying “use cfg.headmodel instead of cfg.hdmfile” but that is hardcoded in the dipfit code. 

 

In this image (https://urldefense.com/v3/__https://imgur.com/P8CvQA3__;!!Mih3wA!W92KXdh7blda6IzoYtjvGjRu6xaTeySMzUMtOXyW_F1oJlZEaDjOLADtJxU5Hp5QVil1qw$ ) you can see the DIPFIT plot with the RV% and TAL numbers:

 

 

I’ve used the same EDF dataset and tested eeglab 2019_0 against 2019_1, and within each of those trials, have tested DIPFIT 3.0 and 3.3, but haven’t been able to find any combination wherein the dipoles actually matched the components, most likely ruling out a version mismatch. 

 

What are we doing wrong? Can anyone shed some light on what might be happening and how we can attempt to further diagnose? 

 

Updates: 

Someone suggested plotting the channels to ensure that channels aren’t being transposed somehow. It appears that the channels are being properly referenced and sequenced. 

This image (https://urldefense.com/v3/__https://imgur.com/ByfaSYi__;!!Mih3wA!W92KXdh7blda6IzoYtjvGjRu6xaTeySMzUMtOXyW_F1oJlZEaDjOLADtJxU5Hp69f1htbg$ ) shows the layout of the channels. Ignoring the pre- and post-fix labels, these appear to be in the correct order and placement in the 10-20 system. These are obtained in the channel locations dialog from the eeglab2019_1/sample_locs/Standard-10-20-Cap19.ced file that I’ve merely changed the labels so I don’t have to manually edit every channel name.  

 

Without going too far down a rabbit hole, when I plot the channel layout obtained from the plugins/dipfit/standard_BEM/elec/standard_1005.elc, I get a plot where the ears (T3/T4) are too low, and the Fp1/Fp2 are too far forward and off the head (https://urldefense.com/v3/__https://imgur.com/dlp8KEw__;!!Mih3wA!W92KXdh7blda6IzoYtjvGjRu6xaTeySMzUMtOXyW_F1oJlZEaDjOLADtJxU5Hp52heQLlw$ ). Is that expected behavior? 

 

 

 Suggestions welcome!

 

Best regards,

Curtis