DeFacing: need for automated wide comparison

[yarikoptic]

The number of defacing and refacing* algorithms grows. Data acquisition changes. New analysis approaches etc.

Publications investigating effects targeting current defacing pipelines on some datasets already exist:

• https://cdn-akamai.6connex.com//645/1827//OHBM2020Poster_15919688426481638.pdf
• https://link.springer.com/content/pdf/10.1007%2Fs00330-019-06459-3.pdf

and similarly to alignment algorithms we don't know which is the best generally/in specific use cases. It would be nice to establish a fully automated pipeline which would grow with newer algorithms and possibly datasets

• could be without knowing ground truth -- just impact (difference) in the results of the pipeline (e.g. simple BET or some other morphometrics) across defacers
• could use some curated dataset(s) with ground truth known as e.g. pipeline + annotated datasets in https://mindboggle.info/ to judge the impact

Anyone interested to join @con/obc ?

• e.g. new refacing in AFNI, See poster 1030 on https://datalad-datasets.github.io/ohbm2020-posters/#/

[SamGuay]

The link to the poster 1030 is broken but can be found on the OHBM poster database at https://ww4.aievolution.com/hbm2001/index.cfm?do=abs.viewAbs&abs=1206.

[DimitriPapadopoulos]

The link to poster 1030-- is broken again, but there's a paper in Frontiers in Psychiatry now: Multisite Comparison of MRI Defacing Software Across Multiple Cohorts

[DimitriPapadopoulos]

My wrong, got it wrong while reading OHBM2020 Posters.

• The above paper relates to poster 1909 Comparison of MRI Defacing Software Across Multiple Cohorts.
• Poster 1030 is Why de-face when you can re-face? by Robert Cox.