Posted By: Andrew Worth - Dec 14, 2015
Tool/Resource: Manually Labeled MRI Brain Scan Database
Neuromorphometrics has just released a new set of manually labeled MRI brain scans -- ready to test your automatic segmentation (labeling) algorithm. We labeled all 40 scans of the The Open Access Series of Imaging Studies (OASIS) project "reliability data". These are 20 nondemented subjects imaged on two visits within 90 days. We comprehensively label the entire brain and parcellate the cortex into regions based on gyral and sulcal landmarks using two neuroanatomical protocols:

1) The Harvard/MGH CMA "General Segmentation"
2) BrainCOLOR cortical parcellation:

We presented a poster giving the Dice Similarity Coefficients (DSCs) at the recent Society for Neuroscience meeting in Chicago. The poster, along with updated plots and a list of min, average, mean, and max DSCs are on our web site: These numbers give a kind of "upper bound" for the reliability of labeling each neuroanatomical region. Some regions are harder to label and/or more variable than others so there is a range of values. You can run your algorithm on the repeat scans and then compare your DSCs to the "best possible" values that can be produced manually. Many authors claim that their results are, "as good as manual." These results demonstrate what manual labeling can do.

OASIS reliability data is from:
We provide the OASIS scans and our labeled results in NIfTI ".nii" format.
RSS Feed Monitor in Slack
Latest News
News archive
Post Comment

This news item currently has no comments.
Post Comment