Some of the more substantive issues listed here are mentioned in the NSD data paper. This list is intended to be comprehensive with respect to all known issues with the NSD dataset (even if very minor). General issues: - Due to scheduling constraints of the participants and the decommissioning of the MR scanner, not all subjects completed all 40 NSD scan sessions. Data quality issues: - Some sessions have substantial head motion (especially subj08) and this will cause artifacts in the data (noise, residual spatial distortion in the pre-processed data). Proceed with caution. - Eyetracking data quality is quite variable due to eye-reflectance issues. - Physiological data acquisition was generally quite good, but occasionally resulted in unusable data. - Some subjects exhibited some drops in behavioral performance (i.e. "bad days"), e.g., subj03-nsd21, subj08-nsd08. - Some sessions may have missing brain coverage due to head displacement. These are deemed "invalid voxels". Regarding fMRI coverage in the NSD core and the 3 additional fMRI sessions (prffloc, nsdsynthetic, nsdimagery), we observe the following: - subj01: no issues - subj02: a few invalid voxels at the inferior tip of the cerebellum in 1 of the 40 NSD core sessions (nsd14) - subj03: no issues - subj04: no issues - subj05: no issues - subj06: for the NSD core sessions, it gets very close at the superior aspect but probably okay, and there are a few missing voxels on the ventral occipital right hemipshere in 1 of the 32 NSD core sessions (nsd04); in prffloc, it gets very close at that same spot (ventral occipital right hemisphere) but probably okay - subj07: no issues - subj08: for the NSD core sessions, it gets very close at the posterior aspect but probably okay, but there is missing a substantial part of the ventral cerebellum in 1 of the 30 NSD core sessions (nsd01); in nsdimagery, there are a few missing frontal voxels; in prffloc, it gets close at the posterior aspect but probably okay - Regarding structural scans, a small portion of the inferior aspect of the cerebellum may cut off in the angiogram (TOF) scans due to limited field-of-view. - In the raw data for the diffusion scans acquired for the 8 NSD subjects, there is some degree of a "striping" artifact that is visible. These stripes alternate every other axial slice and likely reflect an interaction of eddy currents, head motion, and/or scanner vibration. Visually, the artifact looks as if every other axial slice has a different image distortion (e.g. ABABAB... where A is a certain type of distortion and B is a different type of distortion). The magnitude of the artifact varies from subject to subject. In the pre-processing of the diffusion data, this artifact is mitigated using the "slice-to-volume movement" feature of FSL's eddy tool, but be aware that some amount of residual artifact may remain. Minor issues: - Some sessions and subjects may have apparent motion caused by respiration. This is visible in the motion parameter traces (e.g. see subj03-nsd09). - It is possible that some subjects may have pressed buttons while not entirely awake (e.g. subj07). Thus, response rate and reaction time might be imperfect measures of performance. - Some scan sessions had different sets of buttons pressed (e.g. 3/4 instead of 1/2) either due to subject error or due to button box malfunction; this was fixed directly in the raw behavioral files. - Some subjects deliberately pressed a button multiple times (either to ensure that it is detected or as a matter of emphasis). The analysis of the behavioral data recorded the first instance of a string of presses of the same button. - The spatial undistortion process yields some "bleedage" of brain voxels outside of the brain. This generally should not have adversely affected the pre-processing of the data (since such voxels are ignored), but the user should be aware that voxels outside the brain may appear to have signals but these voxels should be ignored. - The nsdsynthetic experiment may have caused afterimages for subjects' perception. - A few structural sessions are missing spin-echo fieldmaps. This is a minor issue as the spin-echo fieldmaps were not used in any pre-processing. Imperfections in experimental data collection: - In the first nsd run of subj07-nsd15, the subject saw about 10 stimulus trials before the run had to be aborted and restarted. - The first nsd run of subj08-nsd28 was accidentally repeated (in entirety) as the second run. The data from this accidental second run was discarded. After the accidental run, the actual second nsd run was conducted. Imperfections in data acquisition: - The recorded reaction times have very small gaps at a rate of about 10 Hz (if you do a histogram with very small bins (i.e. 1-ms bins), you will see the gaps). This is because stimulus frames are shown at a rate of 10 Hz, and during the brief periods of time in which stimulus frames are being shown, inputs from the button box are not polled. Minor error in structural procesing: - Due to a coding error, a double correction was applied in terms of gradunwarp -- the final volumes were sampled from the already-corrected images, thus creating a sort of double correction for the T1 and T2 images. - Fortunately, the logic of the remainder of the processing was preserved. Ultimately, the consequences are at we lose some benefit of the upsampling strategy since it is on the already resampled, and the structural T1 and T2 images are slightly distorted relative to reality. The gradunwarp is a very small effect at 3T (but is more substantial at 7T) (see nsddata/inspections/gradunwarp/). Anomalous brain image features: - Subject subj04 has an incidental non-urgent finding that is visible in the T1 and T2 volumes as well as in the EPI volumes. It is a structure visible in the inferoposterolateral left basal ganglia. It is either an enlarged perivascular space (virchow robin space) or a benign neuroglial cyst. It did not require any clinical follow-up. - Subjects (e.g. subj04, subj06) sometimes have "resetting" motion behavior wherein a drift in head position occurs over the course of a run and then "resets" in the next run. Given our spatial undistortion strategy, this tends to cause subtle shifts in voxel position that are, in a sense, correlated with time. - subj05-nsd03 exhibits some "filling in" in the frontal lobe (less dropout). This is anomalous and does not appear to occur in the other scan sessions. - subj07-nsd36 appears to have a particularly bad shim (especially in ventral aspects of the brain). However, the fieldmap measurements were valid and appear to largely correct the spatial distortions. Nonetheless, the images from this sessions are slightly anomalous compared to the other scan sessions from this subject. - subj08 has a transient sinus infection around nsd12. This shows up as a hyperintensity in the fMRI images around frontal cortex, and persists for a few sessions. The mean signal intensity of nearby voxels is actually improved (brightened) by this effect for the few sessions in which it occurs. Missing behavioral data: - subj04-nsd23 - The 3rd NSD run experienced a "stuck button" during the run. Behavioral responses are therefore missing for this run. - subj06-nsd19 - The 10th NSD run did not have a behavioral .mat file saved due to computer crash. - A simulated replacement .mat file was constructed to allow processing to proceed. - Behavioral responses are therefore missing for this run. - Note that missing behavioral responses are marked by column 19 in the responses.tsv files. Fieldmap anomalies: - subj02-nsd02 - The slice slab for the EPI data was slightly offset from the slice slab for the fieldmaps (by approximately 6 mm). We estimated the offset and adjusted for this in the pre-processing accordingly. - subj03-nsd06 - The 1st fieldmap had poor data. We ignored this fieldmap in the pre-processing. - subj04-nsd11 - The 4th fieldmap had poor data. We ignored this fieldmap in the pre-processing. - subj07-nsd18 - The 4th fieldmap was not collected; we compensated for this in the pre-processing accordingly. Split scan sessions: - This refers to scan sessions that were interrupted and completed at a later time. Some sessions were completed in the same day (e.g. bathroom break), but others were completed many days later. In general, to handle the split scan sessions, we performed the following common steps: (i) fMRI runs and behavioral files from different sessions were collected and placed into single directories, (ii) fieldmap processing was adjusted accordingly, and (iii) behavioral data were concatenated (even if long periods of time elapsed). Note that the ISCORRECTCURRENT values in the responses.tsv files are inaccurate for the split scan sessions in which behavioral trials actually span different days. Also, note that motion parameter estimates may appear to drastically jump for split scan sessions, simply due to the re-positioning of the slice slab for new head positions. - subj01-nsd02 - This session had two parts. The first part consisted of the first three NSD runs. Note that the second NSD run had complete signal loss for a few volumes in the middle of the run. This was dealt with by repeating the 128th volume and using it to fill through the 152nd volume. The second part (completed 7 days later) consisted of the remaining nine NSD runs. - subj06-nsd06 - The subject exited the scanner after NSD run 10 (bathroom break) and then went back in for the remaining runs. New fieldmaps were acquired and used accordingly. - subj06-nsd14 - The subject exited the scanner after NSD run 8 (bathroom break) and then went back in for the remaining runs. New fieldmaps were acquired and used accordingly. - subj06-nsd31 - This session had two parts. The first part consisted of the first two NSD runs. Note that due to scanner malfunction, these two runs only acquired the first 181 TRs (not 188 TRs). To deal with this, in pre-processing, we repeated the 181th volume for the remaining 7 volumes. The second part (completed 12 days later) consisted of the remaining 10 NSD runs. - subj08-nsd02 - This session had two parts. The first part consisted of the first three NSD runs and a part of the fourth NSD run (the run was cut short due to the MR scanner failing). In the second part (completed 7 days later), we resumed the fourth NSD run from where we left off. For the first part of the 4th run, we had 111 usable brain volumes (subsequent acquired volumes were simply deleted). To compensate for the splitting of the 4th NSD run across sessions, we stitched together the behavioral data for the two parts of the 4th NSD run, and we also stitched together the brain volumes for the 4th run (this involved, in the pre-processing, some intensity normalization, appropriate fieldmap handling, and some interpolation of brain volumes). Note that there is, even after pre-processing, some residual abrupt intensity change during the 4th run. Also, note that we compensated appropriately for the discontinuity in the tSNR and FD metrics and the behavioral results corresponding to this 4th run. - subj08-nsd17 - This session had two parts. The first part consisted of the first two NSD runs only (subject had to exit the scanner due to anxiety). The second part (completed 3 months later) consisted of the remaining 10 NSD runs. The behavioral response file reports the correct times. Eyetracking: - The video capture (.mp4) for subj01-nsdimagery run number 9 of 12 is missing about 9-10 seconds of its duration. It is unclear where in the duration this occurred. We return a video capture clip length that is still the correct length (thus, the extends after the end of the actual fMRI acquisition). - The .edf file for the first run of eyetracking in the nsdsynthetic experiment for subject 5 was unexpectedly truncated and is incomplete (nsddata_timeseries/ppdata/subj05/eyedata/eyedata_nsdsynthetic_run01.edf). This run was omitted from the eyetracking data pre-processing. - Subject 5 did exhibit some noticeable levels of eye movements, as reported in the NSD data paper. - Subject 4 exhibited an unusually large number of eye blinks.