McIntosh-Lab / tvb-ukbb

TVB-UKBB Pipeline: TheVirtualBrain implementation of the UK Biobank pipeline
Other
23 stars 12 forks source link

Poor brain extractions in older adult data #71

Closed shen4brains closed 3 years ago

shen4brains commented 3 years ago

Pipeline currently relies on BET for brain extractions; may need alternative brain extraction techniques to achieve better performance in older adult data. Example subjects with poor brain extractions in camCAN dataset include: sub-CC120987, sub-CC420241, sub-CC512003, sub-CC720497 and sub-CC723395

Relevant code here: https://github.com/McIntosh-Lab-RRI/tvb-ukbb/blob/35483bfcafb52db9a510afd18bada12203653c78/bb_structural_pipeline/bb_struct_init#L46-L47

shen4brains commented 3 years ago

Also poor brain extractions in dMRI data (see pptx in Slack); relevant dMRI extractions occur here: https://github.com/McIntosh-Lab-RRI/tvb-ukbb/blob/35483bfcafb52db9a510afd18bada12203653c78/bb_diffusion_pipeline/bb_eddy/bb_pre_eddy#L46

and here: https://github.com/McIntosh-Lab-RRI/tvb-ukbb/blob/66749cc43a3b139a3d92e144f562c39306342046/bb_diffusion_pipeline/bb_eddy/bb_post_eddy#L31

shen4brains commented 3 years ago

dMRI brain extraction issue may be specific to camCAN; as of yet haven't seen this consistently in ADNI

shen4brains commented 3 years ago

T1 brain extraction is sometimes poor in ADNI3

shen4brains commented 3 years ago

Original issue not due to poor brain extractions but to the earlier failure of nonlinear registrations; Opened new issue #76 to address nonlinear registrations. Keeping this issue open as it still applies to ADNI3 T1 data, and maybe camCAN dMRI data as well

shen4brains commented 3 years ago

Final decision re: 1) ADNI3 T1 brain extractions: not improved by optiBET (cf @shen4brains' local branch add_optiBET) so T1 brain extractions will remain as-is 2) cam-CAN dMRI B0 brain extractions improved by adopting T1-style brain extractions (applying MNI_brain via warp to native space) and implemented across all branches (master, cam-CAN, ADNI3) for consistency