hide
Free keywords:
-
Abstract:
Automatic image segmentation at magnetic resonance imaging (MRI) of the brain is
essential for a number of applications. Many well-known segmentation tools exist for the
clinical domain. However, we have found that they become unreliable when applied to
ultra-high resolution images and, in particular, to data acquired at magnetic field strength
of 9.4 T. This has motivated us to develop a segmentation method that can handle images
at ultra-high resolution ≤ 0.6 mm and field strengths 1.5–9.4 T. Specifically, we propose
an adversarial game for flexible domain adaptation of convolutional neural networks in the
context of brain MRI segmentation. In particular, we develop FLEXseg, the first brain
MRI segmentation method suitable for images acquired at 9.4 T with 0.6 mm isotropic
resolution. We demonstrate the performance of FLEXseg by comparing it with manually
corrected labels approved by expert neuroradiologists.
