Keywords:
Artificial Intelligence, Anatomy, Neuroradiology brain, MR, Segmentation, Computer Applications-General, Computer Applications-Detection, diagnosis, Dementia
Authors:
I. Evangelista, C. L. Galimberti, G. Pascariello, J. C. Gomez, A. L. Rodríguez Musso, P. Donnelly-Kehoe; Rosario/AR
DOI:
10.26044/ecr2019/C-2108
Conclusion
FreeSurfer’s morphometric measurements are influenced by the MRI acquisition parameters.
However,
we introduced a method to remove those effects by using standardization techniques,
allowing the translation of automatic quantitative MRI analysis from research to clinical applications.
We developed a technique for harmonizing data extracted from sMRI acquired using different field strengths and measured its effectiveness in terms of the ability to confound a random forest classifier.
Our method consisted in calculating a 10-year-bin z-score for each class (1.5 T and 3.0 T).
Our approach succeeds in reducing class separation without jeopardizing a classifier's ability to detect diseases.
Harmonization between field strength groups makes it easier to compare images from different sources,
albeit there are a number of variables that cannot be used indistinctly.
Since it is important to account for these features,
an analysis of feature importance was conducted in order to determine which variables should not be analyzed indistinctly in spite of the harmonization process.
Finally our approach did not affect the classifier’s capacity to differentiate AD,
MCI and HC.