Abstract
Multisite MRI-based machine-learning benefits from larger sample sizes, but the use of different MRI devices introduces effects of the site (EoS). Several methods (e.g., ComBat) remove EoS, increasing accuracy. However, these methods may leave residual EoS that biasedly inflate the accuracy. Thus, we cannot know if EoS-removal methods associated with increased accuracy truly remove more EoS or, conversely, fail to remove EoS that inflate the accuracy. We present a strategy to measure the true efficacy of EoS-removal methods.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have