Challenge
- The size of the HBN makes expert QC prohibitve for the full dataset.
- Likewise for future releases.
Solution
Use community scientists to amplify expert QC ratings
-
Based on the SwipesForScience
framework, we created a web-app that presents views of b=0 and
directionally-encoded FA maps for a binary (pass/fail) decision for
1,653 HBN subjects.
-
374 community scientists provided 587,778 ratings for a mean of > 200
ratings per subject.
-
133 new co-authors!
Amplifying expert scores from Fibr raters
-
We trained a gradient boosting model to predict expert scores based on
both Fibr ratings and automated QSIPrep metrics.
-
This outperformed a model trained on only the QSIPrep metrics,
demonstrating the added value of the community science raters.