Neuro MRI annotation with post-op workflow and QC iteration
Annotation of brain MRI following an iterating protocol, including post-operative site handling, radiology-report QC and granular artifact classification.
A neuroimaging AI vendor was rolling out a new version of its annotation protocol that changed the handling of post-operative cases, added a radiological-report quality check and introduced finer-grained artifact categories. Annotators had to absorb the changes without regressing on the existing pathology-level rules.
Annotate brain MRI studies under the updated protocol: annotate post-op sites as an additional finding (with pathologies outside the surgical bed segmented as usual), classify report quality (match / patient mismatch / not readable) and select artifact categories (blurry, ghosting, fold-over, metal, zipper, etc.).
The annotation team was re-trained on the version delta, with decision trees for wrong-sequence cases (Apollo-flagged MIP as SWI, etc.) and clear rules for when to exclude vs. annotate. Radiologists systematically applied the report-QC selection and used the updated masking settings on every case.
The client absorbed the protocol update without dataset drift: post-op cases were kept in the dataset with correctly localized findings, and the finer artifact taxonomy made downstream data curation faster.