Harmonizing Brain MRI for Migraine Classification
Author Information
Author(s): Yoon Hyunsoo, Schwedt Todd J., Chong Catherine D., Olatunde Oyekanmi, Wu Teresa
Primary Institution: Mayo Clinic
Hypothesis
Using a healthy core can improve the generalization of machine learning models for classifying migraine.
Conclusion
The study demonstrates that incorporating a healthy core significantly enhances the accuracy and generalizability of migraine classification models.
Supporting Evidence
- Using a healthy core improved classification accuracy by 25%.
- Multicenter studies may introduce confounding factors affecting model generalizability.
- Data harmonization strategies can mitigate site-specific biases.
Takeaway
The researchers found that using a specific group of healthy individuals helped improve the accuracy of identifying migraines using brain scans.
Methodology
The study utilized two datasets and employed Maximum Mean Discrepancy in Geodesic Flow Kernel space to harmonize MRI data.
Potential Biases
Potential biases due to participant characteristics and differences in imaging protocols across centers.
Limitations
The study only partially describes the heterogeneity of the healthy control cohort and relies on non-imaging data.
Participant Demographics
Participants included 120 individuals (66 with migraine, 54 healthy controls) and 76 individuals (34 with migraine, 42 healthy controls), aged 18-65.
Statistical Information
P-Value
p<0.01
Statistical Significance
p<0.01
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website