jcohenadad
commented
7 months ago Thank you for the detailed figure layouts-- it makes it easier to comment.
- TFL_B1_C3C4:
- B1+: I would not average per slice and then compute the SD across slices, because the mean is sensitive to outliers. Instead, I would simply compute the SD across all voxels in the ROI (ie: SC mask), and 'play around' with ways to mitigate the presence of outliers (eg: remove 'noisy' voxels from the ROI before computing the SD-- we need to find a good way to remove those voxels-- playing the the code and the data, I will get better ideas how to do it).
- I find it more informative to show the evolution of B1+ along the cord vs. the efficiency. So for B1+ I would have a panel A showing the actual maps for a representative subject, panel B showing the slice-wise mean across the SC for that subject, and panel C showing the SD for each subject. Alternatively, for panel C, we can consider showing the violin plot of B1+, and report the SD in a table instead.
- efficiency: I would just represent the mean efficiency inside the ROI, for each subject and each site (again, making sure 'noisy' voxels are not included)
- COILQA_SAG_LARGE
- SNR: People really like to see SNR maps, so I would also show one for a representative subject, and then show the mean SNR in the ROI along the cord for one subject, and the mean for all subjects (eg: the same panels A B and C as suggested above for B1+). That way people will also be able to assess uniformity.
- g-factor:
- People commonly look at g-factor maps across various acceleration factors, because it gives a better sense of where g-factor is increased in the image, how the pattern looks like, etc. I would not reduce the information with numbers in an ROI. So I would just pick a representative subject and show maps of R=2x2, 2x3, 3x2, 3x3
- Another consideration about the gfactor: depending how "large" the FOV is, I would not show the g-factor if there is too much background along the phase encoding direction. So I would just pick the most appropriate g-factor maps and show them in the paper. Happy to help pick the most appropriate one (we can open another issue and show all the acquisitions, and we can brainstorm there)
- correlation matrix: I would not show these-- they are too 'low level' and are usually more a quality assessment of a given coil, rather than a performance assessment (which is what this paper is about).
- DREAM_LARGE: same as the B1+ fig with panel A/B/C
- DREAM_MEDIUM + DREAM_MEDIUM_066 + DREAM_MEDIUM_HWLIMIT:
- I would go with fig 5 (no violin plot-- it becomes too overwhelming)
- GRE
- I think that the tiled figure with each individual channel will be overwhelming (to make and to look at), and also this is quite a 'low level' assessment. The bigger picture is assessed with the distribution of SNR along the cord
- Instead, I would show representative images of T2* across sites (eg: one site per column, one level per row)
evaalonsoortiz
Our dataset incudes 7 sites with 3 subjects per-site (the same 3 subjects were, or are to be, scanned across all 7 sites). The goal is to compare coil designs. The purpose of this issue is to develop a consensus on what metrics we want to derive from the multi-subject data and how we want to visualize those metrics across sites.
Please comment on whether or not you agree with the "goals" and the proposed figures. Alternatively, please suggest and "goal" and/or figure for a given scan type.
Images:
TFL_B1_C3C4 ---> B1+ map acquired with the Optimal Reference Voltage
Goals:
Suggested figures: Figure 1:
Figure 1 shows us how uniform B1+ is along the SC but it does not tell us how efficient the coil is. For that we may want to include the following figure:
Figure 2:
COILQA_SAG_LARGE (H>F phase encoding)
Images: SNR, g-factor, correlation matrix
Goals:
Suggested figures:
For SNR: same as those for TFL (B1+)
For g-factor:
Using an enlarged spinal cord (SC) ROI, compute the average g-factor along the SC report it along with the standard deviation.
Figure 3:
The drawback of Figure 3 is that the little bars for each subject might overlap quite a bit making it overall hard to read. Alternatively we can represent the g-factor along the SC for each subject as a histogram/distribution in the form of a ridge-plot:
Figure 4:
Figures 3 or 4 can be a sub-plot for a larger figure that includes other acceleration factors (3x3, 3x2).
For correlation matrices:
Compute the average correlation matrix across subjects for each site and show the average matrix for each of the 7 sites in a tiled arrangement.
COILQA_SAG_SMALL (H>F phase encoding)
Goals:
Suggested figures:
Same as Figure 3 or 4. This could be included in supporting documents.
COILQA_TRA (L>R phase encoding)
Goals:
Suggested figures:
For SNR: Compute in-plane standard-deviation and average that for all slices. Show this value for each subject, and across sites, like in Figure 1.
For g-factor: Same as Figure 3 or 4.
DREAM_LARGE
Goals:
Suggested figures:
Same as Figure 2. Note: A bias in measured B1+ where the refV is not well calibrated can also lead to differences in measured B1+ uniformity. Therefore we may want to have another version of Figure 1.
DREAM_MEDIUM + DREAM_MEDIUM_066 + DREAM_MEDIUM_HWLIMIT
Goals:
Suggested figures:
Compute the average B1+ along the spinal cord and average across subjects, for each refV. The former will be included in a tiled figure where each sub-plot represents one site.
Figure 5:
Each site has 3 violin plots (refV, 0.66refV, 1.5refV) containing the distribution of B1+ values for all subjects.
Figure 6:
GRE ---> uncombined GRE images
Goals:
Suggested figures:
A tiled figure (containing each channel's GRE magnitude) could be generated for each subject at each site and presented in supporting documentation.