Unaligned labeling of discs for T2w images after sct_label_vertebrae in process_data.sh

[sandrinebedard]

Description

While running the pipeline on 30 subjects, 8 out of 30 subjects have disc labeling not aligned with the spinal cord for T2w images. In process_data.sh, disc labeling is generated with T1w image and then the template is use for T2w with sct_register_multimodal. This suggest that the patient moved between acquisitions.

The results were obtained by running the following line on Joplin:

sct_run_batch -jobs -1 -path-data ~/duke/temp/sebeda/test1_30_sub/data_BIDS -path-output ~/ukbiobank_results/ -script process_data.sh -script-args $PATH_GRADCORR_FILE

The results of the test are now in: duke/temp/sebeda/test1_30_sub/ukbiobank_results The following subject present this problem:

• sub-1000710
• sub-1000537
• sub-1000710
• sub-1000918
• sub-1000985
• sub-1002169
• sub-1002191
• sub-1002721

Example of the problem

For the subject sub-1000710, QC for sct_label_vertabrae for T1w:

QC for sct_label_vertabrae for T2w: While viewing images sub-1000710_T1w.nii.gz and sub-1000710_T2w.nii.gz, we can see that the patient moved between the two images:

Maybe generate a label template for discs with sct_label_vertebrae for T2w as done for T1w.

[jcohenadad]

Vertebral labeling is only required for finding the slices associated with specific vertebral levels (e.g. C2, C3), subsequently used for aggregating CSA results.

So in the example you showed above, the patient slightly tilted the head, which resulted in a misalignment of the labeled segmentation, however the "Z" corresponding to the vertebral labeling is still somewhat correct, therefore I would not worry about this.

It is good to output and verify the QC as you did here, and in case there is a large movement in the superior-inferior direction between the T1 and T2 scans (resulting in a wrong estimation of level on the T2 scan), then we can add a manual labeling step for those subjects. I anticipate this will concern <5% of the population, so it will likely represent 40,000 0.05 10s / 3600 ~ 6h of labeling work (which we could split across 6 people, so 1h each).

[sandrinebedard]

Vertebral labeling is only required for finding the slices associated with specific vertebral levels (e.g. C2, C3), subsequently used for aggregating CSA results.

So in the example you showed above, the patient slightly tilted the head, which resulted in a misalignment of the labeled segmentation, however the "Z" corresponding to the vertebral labeling is still somewhat correct, therefore I would not worry about this.

It is good to output and verify the QC as you did here, and in case there is a large movement in the superior-inferior direction between the T1 and T2 scans (resulting in a wrong estimation of level on the T2 scan), then we can add a manual labeling step for those subjects. I anticipate this will concern <5% of the population, so it will likely represent 40,000 0.05 10s / 3600 ~ 6h of labeling work (which we could split across 6 people, so 1h each).

Perfect in that case! Thank you