No BOLD components detected! Please check data and results!

Software Support
, , , ,
1

Dear experts,
@e.urunuela @dowdlelt

I am seeking assistance with a persistent error occurring in a subset of subjects during our task-based fMRI preprocessing pipeline using AFNI’s afni_proc.py and multi-echo TE-dependence analysis (tedana).

The Problem

For a number of subjects, the tedana component returns the critical warning/error: “No BOLD components detected!”

  • Other subjects run successfully without this issue.
  • I have verified that the subjects with the error do not exhibit unusually high motion compared to the successful subjects.

Acquisition Details

  • Field Strength: 7 Tesla
  • Voxel Size: 1.5 mm isotropic
  • Analysis: Task-based fMRI

Processing Details

I am calling tedana via afni_proc.py using the following option:
-combine_method m_tedana

The resulting generated tedana command, which is failing for these subjects, is:
tedana -d ${d1} ${d2} ${d3} -e 14.0 37.87 61.74 --mask ${mask_epi_anat} --maxit 5000 --tedpca kundu-stabilize --fittype curvefit --tree meica --n-threads 30 --ica_method robustica --n_robust_runs 30 --maxrestart 500 --lowmem --gscontrol mir gsr --out-dir tedana_r01 --convention orig

Troubleshooting Attempted

I have tried the following variations, but the error persists for the specific subset of subjects:

  • Changing --tedpca options to “AIC” and “KIC”.
  • Changing --tree to “tedana_orig”.

Example Data

I uploaded an example run (and BIDS .json for the echo1 dataset)that exhibits this error here: https://www.dropbox.com/scl/fo/439bttcbln35ywk9xlnqo/AMIVcj6P4OIFi_uRhfndCEM?rlkey=m0ti3pqf7t1ws0nssq9kw83fr&st=vvsj7q69&dl=0

Here is the afni_proc.py command:

afni_proc.py -subj_id 816 -scr_overwrite 
-dsets_me_run ${run1echo1} ${run1echo2} ${run1echo3}                                                                 
-dsets_me_run ${run2echo1} ${run2echo2} ${run2echo3}                                                                                                                
-dsets_me_run ${run3echo1} ${run3echo2} ${run3echo3}                                                                                                             -dsets_me_run ${run4echo1} ${run4echo2} ${run4echo3}
 -echo_times 14.0 37.87 61.74 -out_dir   ${outdir}                                                                                        
-blocks despike tshift align tlrc volreg mask combine blur scale                                                                
regress -blip_forward_dset ${AP} 
-blip_reverse_dset ${PA}                                                                                                                                                    -blur_size 3 
-blur_in_mask yes 
-copy_anat  T1_bc_SS.nii.gz  -anat_has_skull no                                                                                                                                           -mask_opts_automask -clfrac 0.3 -dilate 1 
-mask_epi_anat yes
-tshift_opts_ts -tpattern @SliceTiming.1D                                                                                   -align_unifize_epi local 
-align_opts_aea -cost lpc+ZZ -giant_move -check_flip 
-epi_strip 3dAutomask 
-tlrc_base MNI152_2009_template_SSW.nii.gz \
-tlrc_NL_warp \
-tlrc_NL_warped_dsets anatQQ.816.nii \                                                                                                                                           anatQQ.816.aff12.1D \
anatQQ.816_WARP.nii \
-volreg_align_to MIN_OUTLIER -volreg_align_e2a -volreg_tlrc_warp \                                                         -regress_apply_mot_types demean deriv -regress_motion_per_run \                                                               -regress_censor_motion 0.3 -regress_censor_outliers 0.15 \ 
-regress_stim_times
{regress options}                                                               
-combine_method m_tedana  
-combine_tedana_path /hpc/users/chowda26/.local/bin/tedana 
-combine_opts_tedwrap -tedana_is_exec 
-combine_opts_tedana '--maxit  5000 --tedpca kundu-stabilize 
--fittype curvefit --tree meica --n-threads 30 
--ica_method robustica --n_robust_runs 30 
--maxrestart 500 --lowmem  --gscontrol {mir,gsr}'

Any insight into why tedana might be failing to detect BOLD components only in this subset of 7T data would be greatly appreciated.

Thank you,
Avi

2

Hi, Avi-

I will defer to the tedana experts you referenced here for your main question of interest. I do have a couple quick comments about your afni_proc.py (AP) command, please see below.

Also, I put copies of your posted AP and tedana commands with some vertical alignment formatting for options, just to make it easier to read for myself and potentially others.

As a general question here, did this complete processing enough to create the APQC HTML report? I’m guessing perhaps not, if tedana failed during the combine block of processing. It might be worth running this participant’s data with just optimal combination (-combine_method OC) to get through processing and see if anything else is going awry in processing—too much motion, poor signal, artifact, alignment failing, or some other woe. (Note, several pieces to check are already in the results directory, but many haven’t been put together before regress, such as a total censor file. But you could go through those, too.)

And certainly, the others you pinged might be able to troubleshoot this tedana output more directly. Anyways, further notes and formatted commands below.

–pt

AP command, reformatted with vertical alignment:

afni_proc.py                                                                 \
    -subj_id                  816                                            \
    -scr_overwrite                                                           \
    -out_dir                  ${outdir}                                      \
    -dsets_me_run             ${run1echo1} ${run1echo2} ${run1echo3}         \
    -dsets_me_run             ${run2echo1} ${run2echo2} ${run2echo3}         \
    -dsets_me_run             ${run3echo1} ${run3echo2} ${run3echo3}         \
    -dsets_me_run             ${run4echo1} ${run4echo2} ${run4echo3}         \
    -echo_times               14.0 37.87 61.74                               \
    -copy_anat                T1_bc_SS.nii.gz                                \
    -anat_has_skull           no                                             \
    -blocks                   despike tshift align tlrc volreg mask combine  \
                              blur scale regress                             \
    -tshift_opts_ts           -tpattern @SliceTiming.1D                      \
    -blip_forward_dset        ${AP}                                          \
    -blip_reverse_dset        ${PA}                                          \
    -align_unifize_epi        local                                          \
    -align_opts_aea           -cost lpc+ZZ                                   \
                              -giant_move                                    \
                              -check_flip                                    \
                              -epi_strip 3dAutomask                          \
    -tlrc_base                MNI152_2009_template_SSW.nii.gz                \
    -tlrc_NL_warp                                                            \
    -tlrc_NL_warped_dsets     anatQQ.816.nii anatQQ.816.aff12.1D             \
                              anatQQ.816_WARP.nii                            \
    -volreg_align_to          MIN_OUTLIER                                    \
    -volreg_align_e2a                                                        \
    -volreg_tlrc_warp                                                        \
    -mask_opts_automask       -clfrac 0.3                                    \
                              -dilate 1                                      \
    -mask_epi_anat            yes                                            \
    -blur_size                3                                              \
    -blur_in_mask             yes                                            \
    -regress_apply_mot_types  demean deriv                                   \
    -regress_motion_per_run                                                  \
    -regress_censor_motion    0.3                                            \
    -regress_censor_outliers  0.15                                           \
    -regress_stim_times       {regress options}                              \
    -combine_method           m_tedana                                       \
    -combine_tedana_path      /hpc/users/chowda26/.local/bin/tedana          \
    -combine_opts_tedwrap     -tedana_is_exec                                \
    -combine_opts_tedana      '--maxit 5000 --tedpca kundu-stabilize --fittype curvefit --tree meica --n-threads 30 --ica_method robustica --n_robust_runs 30 --maxrestart 500 --lowmem --gscontrol {mir,gsr}'

One thing I noticed here after reformatting was that the {regress_options} argument appeared to be a shell variable that is missing a $ to the left of the first curly bracket. I am not sure if that is exactly what your code has, or whether that was something introduced when posting here, but that would be a problem with running the AP command.

Another thing I might note is that I’m not sure despiking would be recommended for task-based analysis, esp. with event related tasks (if that is what you have). A strong, short response could look like a spike, and then be inappropriately reduced prior to regression modeling.

Similarly, I’m not sure that for task analysis I would generally include this:

-regress_apply_mot_types  demean deriv

… because usally the deriv is added primarily for resting state analyses, as trying to remove a bit extra motion effect, if possible. For task-based, I don’t think that would be necessary, and it would reduce degrees of freedom unnecessarily. Esp. as you have ME-FMRI data to combine here, too.

For QC purposes, you might consider adding these options, to have additional useful outputs added to the APQC HTML (see here and here for more discussion/description of these):

-radial_correlate_blocks  tcat volreg regress
-volreg_compute_tsnr      yes
-regress_make_ideal_sum   sum_ideal.1D

Here is the tedana command reformatted a bit, as well.

tedana                                                                       \
    -d  ${d1} ${d2} ${d3}                                                    \
    -e  14.0 37.87 61.74                                                     \
    --mask     ${mask_epi_anat}                                              \
    --maxit    5000                                                          \
    --tedpca   kundu-stabilize                                               \
    --fittype  curvefit                                                      \
    --tree     meica                                                         \
    --n-threads   30                                                         \
    --ica_method  robustica                                                  \
    --n_robust_runs 30                                                       \
    --maxrestart  500                                                        \
    --lowmem                                                                 \
    --gscontrol   mir gsr                                                    \
    --out-dir     tedana_r01                                                 \
    --convention  orig
3

Thank you so much @ptaylor for your input and for reformatting the command.

To clarify: I didn’t want to copy-paste all the timing files here, so I just wrote {regress_options} as a placeholder—it is not an error in my actual script.

Regarding the “No bold components” warning: All of the subject runs where this warning appeared still fully completed processing. Even though no components were explicitly identified as BOLD, tedana proceeded by accepting components that were not categorized as BOLD but were also not rejected (i.e., the “accept” list).

I will try again with your recommended settings: -regress_apply_mot_types demean only and removing despiking from the block. However, I wanted to note that the regress option is usually applied after tedana, so I am unsure if that change will influence the tedana preprocessing steps.

Thank you, Avi

4

Hi, Avi-

OK, sounds good about the {regress_options} syntax there.

So, if that processing ran to completion, you should have the afni_proc.py QC HTML output, which even includes a button to jump to the tedana QC HTML pages from there. How do the various processing outputs look? Can you verify that alignment was all good, that there aren’t obvious issues apparent with the data, that motion is reasonable, etc.? Also, for the benefit of troubleshooting the tedana command, it might be worthwhile posting a screenshot of the tedana HTML page here, perhaps.

–pt

5

Yes, I do have the QC HTML output. Alignment to MNI looks OK to me, and motion is not particularly high compared to other subjects where BOLD components were detected. Signal dropout in ventral regions is also not notably different for these subjects.

Looking at the tedana QC HTML, it is unclear to me why specific components that have high kappa but low rho were still not accepted as BOLD. The kappa threshold seems to be high for these subjects.

Here is a snapshot of an example component:

image
image1920×951 113 KB

6

For what it’s worth, I don’t think any of the current tedana developers (myself included) regularly use, or have a ton of confidence in, the gscontrol methods. The “gsr” option will affect the data that’s being run through ICA.

1 Like