Date written: 2020-05-26
Last ran: 2020-05-26

Description: Here, we review one SPINS participant, SPN01_CMH_0135 in detail, at every step along processing, from (1) raw data preprocessing, to (2) Slicer tractography, to (3) use of the ORG atlas. This participant is a healthy control and was randomly chosen from those who passed QC and had low in-scanner motion.

Links:
Review 1: Preprocessing
Review 2: Slicer tractography
Review 3: ORG atlas

[1] Raw data

Description. There’s no missing slices / major artifacts, and we have all 60 directions. The mean absolute relative motion across the entire DWI acquisition is .064mm (maximum 0.087) and mean relative motion is 0.072mm (maximum 0.099). There is no special comment about participant behaviour in REDCap, and no special notes about data quality on the dashboard QC.

raw data

tissue segmentation

directions

[2] Preprocessing

Description. The script used to preprocess SPINS data, as presented here, can be found here. Note: Michael replaced the b0_avg, be_pre and bet_pre_mask steps with the outputs from the dmriprep b0 reference workflow. Previous iterations of SPINS preprocessing can be found in datman. Though we continue to see some anterior distortion (see axial slice), this is much improved compared both to uncorrected data, and other software applications.

[a] Eddy correction

Description. We used eddy to correct for eddy current-induced distortion and subject movement (but not distortion correction). Review of the eddy quad report, which shows individual subject measures, suggests good quality data.
Special note. The preprocessing of this data does not include denoising, as the HCP pipeline does not include denoising, and our preprocessing was meant to be similar to HCP at one point. However, Michael has done a qualitative review of SPINS-ASD data and recommends denoising of this sequence.

tSNR (b=0)

CNR (b=1000)

outliers

[b] Distortion correction

Description. We correct for distortion using the BrainSuite Diffusion Pipeline (BDP), which corrects via registration (cf. fieldmaps). More specifically, the software “uses a constrained non-rigid registration based on mutual-information, and uses the bias-field corrected anatomical image generated by BrainSuite as a registration template and constrains the registration using spatial regularization and physics-based characteristics of distortion in EPI sequences.”

[c] Brain masking

Description. Brainmasking was conducted using a combination of BET and AFNI. A prior report of what motivated this decision can be found here. The masks look good! (Note that we have opted to use Slicer’s native brain masks for tractography.)

[3] Data manipulation


[a] Remove floats

Description. Eddy creates non-integer (float) values. Though Slicer allows for this, we wanted to create images without floats to allow for closer comparison to other pipelines, such as dtifit. As expected, removing floats (via FSL) changes the minimum and maximum contrast values change from floats to integers (min: -497.279 to -497; max: 30457.54 to 30457). The datatype, as viewed via the FSLeyes ‘overlay information’ panel changes from FLOAT32 to INT16. The code for this process can be reviewed here.

[b] Create .nrrds

Description. We used Slicer to convert the diffusion data, as well as the b0, to .nrrd (cf. nii) format, as is required by Slicer. The raw data remain quantitatively identical. The code for this process can be reviewed here.

[4] Conclusion


We think that everything looks good with the preprocessing (though will proceed to de-noise), and, if time allows, test scdflows for further improvements to distortion correction.