Abstract:
Standard Magnetic Resonance Imaging (MRI) sequences such as T1-MRI’s are useful
for brain structure characterisation but lack information about the vasculature in
the brain. Recovering vasculature information requires the use of non-standard sequences
such as Time of Flight (ToF) Magnetic Resonance Angiography (MRA), and
its scarcity in medical databases hinders the retrospective haemodynamic studies of
the brain.
The Generative Adversarial Network (GAN) is a recently developed form of
deep learning that has shown potential to transform images from one image domain
into another. This work investigates the use of GAN techniques to recover vascular
information from T1-MRI priors by performing a domain transformation into
ToF-MRA. Changes in the vasculature in the brain are correlated with neurological
disorders such as Alzheimer’s Disease, multiple sclerosis, and epilepsy. Increasing
the availability of vascular data will contribute to a further understanding of the
vascular remodelling due to the pathology potentially defining novel biomarkers
for their earlier detection.
Thus, a reliable transformation of this kind would expand the information obtained
in a limited time in a clinical setting and also allow the use of large available
datasets of standard MRI sequences for haemodynamic and vascular analysis.