Automatic brain vessel segmentation using 3D convolutional neural networks

Aims
Design an algorithm to segment vessels from brain MR images using a neural network. Use the vessel segmentation to guide trajectory planning in the implantation of stereo-electroencephalography (SEEG) electrodes in epilepsy surgery.
Background
In a third of individuals with epilepsy, antiepileptic drugs do not control seizures. Of these patients, half have focal epilepsy which may be treated by curative resection if the seizure onset zone (SOZ) has a focal location and is in a part of the brain that may be removed without causing deficits. To locate the SOZ, a number of pre-operative imaging scans (such as T1-weighted MRI, T2* MRI) are acquired. If MRI scans are negative, i.e. no distinct lesions are present, intracranial SEEG electrodes can be implanted to locate the SOZ.
Intracranial haemorrhage is the most common complication associated with electrode implantation. Hence, brain vessels are among the most critical brain anatomy that needs to be assessed for mitigating surgical risks. Automatic, accurate and robust segmentation of vessels is of great interest for clinicians who use computer-assisted planning (CAP) for the implantation of SEEG electrodes.
Recently, neural networks (NN) have proved to outperform traditional processing methods in many medical imaging applications. The main aim of the project is to implement an algorithm that uses deep learning (DL) to perform segmentation of brain vessels on MRI angiograms and venograms.