Explained variance of inverse solution is extremely low for only one subject

💬 Support & Discussions
, ,
1
  • MNE version: 1.4.0
  • operating system: Windows-10-10.0.19045-SP0

I am computing inverse solution (method dSPM) on several subjects. I have done with 4 subjects, their explained variance are all around 70-80%, however , there is one subject, whose inverse solution 's explained variance is only 2-3%. But the codes are all the same, and i have check that the T1 template was right for the subject.

the code return information looked like these :

Applying inverse operator to "1"...
    Picked 306 channels from the data
    Computing inverse...
    Eigenleads need to be weighted ...
    Computing residual...
    Explained   2.3% variance
    Combining the current components...
    dSPM...
[done]

The codes are below:

evoked = epochs.average(by_event_type = True)
noise_cov = mne.compute_covariance(epochs, tmax=0.0, method=["shrunk", "empirical"], rank='info', verbose=True)
singleevent = 0 # list start from 0 
subjects_dir =r"G:\MEG_analysis\sub_mni_freesurfer"
##  inverse      
inverse_operator = make_inverse_operator(
    evoked[singleevent].info, fwd, noise_cov, loose=0.2, depth=0.8
)
method = "dSPM"
snr = 3.0
lambda2 = 1.0 / snr**2
stc, residual = apply_inverse(
    evoked[singleevent],
    inverse_operator,
    lambda2,
    method=method,
    pick_ori=None,
    return_residual=True,
    verbose=True,
)

what 's more, when i want to morph the subject’data to fsaverage , it return error:
ValueError: dimension mismatch.

the code is like these :

##  morph subject surface to group surface
        morph = mne.compute_source_morph(
            stc,
            subject_from=subject,
            spacing = 5,
            subject_to="fsaverage",
            subjects_dir=subjects_dir,
        )
        stc_fsaverage = morph.apply(stc)
2

Hi @ling,

have you plotted the source space results? Do they look plausible?
Do the sensor-level data of that participant look alright? Are they maybe extremely noisy?
Another suspect could be the forward model - you could inspect that if you have not done that yet:
https://mne.tools/stable/auto_tutorials/forward/30_forward.html#sphx-glr-auto-tutorials-forward-30-forward-py
https://mne.tools/stable/auto_examples/forward/forward_sensitivity_maps.html#ex-sensitivity-maps
These two tutorials can help you if you are not sure how to do that and which ingredients to inspect.

HTH,
Britta

2 Likes