Description. Here, we review the success of tractography on the SPINS PI travelling human subject data (shared with Dr. Rathi). We see that, like in the participant dataset, missing tracts are driven by scanner type, with more tracts missing on Siemens vs. GE scanners. Our preprocessing was performed with an in-house variant of dwiprep
. Tractography is UKF (default settings), and mapped to the ORG atlas.
Dataset. The travelling subject dataset is comprised of 4 subjects, from a maximum of 4 timepoints, across a maximum of 6 scanners (CMH = GE Discovery, MRC = Siemens Tim Trio, ZHH = GE Signa; CMP, MRP, and ZHP = Siemens Prisma). Not all travelling subjects were scanned on every scanner or each year. In total, the dataset contains contains a total of 30 scans.
Plot: Missing tracts.
The visualization below shows number of tracts segmented by (i) site and (ii) scanner type.
Table: Fiber summary.
In the sortable table below, the count
variable indicates the number of participants with data for a given tract, and percent
indicates corresponding percentage. In total, we have data for 2075 tracts out of a possible maximum of 2160, i.e., 96.0648148%. However, we see that several individual tracts have segmentation rates in the 60-85% percent range, which is far lower than expected.
Visualization: Complete missingness.
Below is a visual summary of missing data summarized above, by participant. Black indicates that the tract is missing entirely for the given participant. Note: any tract that has at least one fibre is included here. Interestingly, we see that different tracts appear to be missing values, compared to our study sample (however, our sample here may be too small to read too much into that difference).
Table: Missing tracts by participant.
This table complements the image above: here, we summarize how many participants were missing the given number of tracts.
Tracts missing | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 12 | 16 |
Percent missing | 0.00 | 1.39 | 2.78 | 4.17 | 5.56 | 6.94 | 8.33 | 16.67 | 22.22 |
Participant count | 15 | 1 | 3 | 2 | 3 | 2 | 1 | 1 | 2 |
Visualization: Tractography 'quality'
In addition to examining missing values, we review tractography 'quality', i.e., the number and length of segmented fibers. The following plots show, per tract, (i) the number of fibers segmented, and (ii) mean length, for the n=30 unique combinations of participant, scanner, and timepoint, separated by hemisphere. Participant is indicated by colour. Scanner is indicated by geom shape. Timepoint is on the X axis. Lines connect the same participant and same scanner. (Some lines are not connected as not all participants were scanned on all years, or on all scanners.) We see some variability across timepoints, and within scanner. We also see some low values, of both number of fibers and mean length.
Summary. We take it that we aren't getting good tractography results. We aren't sure if the issue is our tractography parameters, our preprocessing, or our acquisition. Dr. Rathi et al.'s analysis on the same data will help us identify the source of our issue.