A cerebral aneurysm is a blood vessel anomaly (abnormality) in the brain where the blood vessels are dilated (widened) (see Fig 1a). It presents a great health risk because it has a high probability of hemorrhage (it can burst and cause stroke, i.e., bleeding of brain blood vessels). Another such anomaly is a cerebral Arteriovenous Malformation (AVM), which takes place at the direct connection between arteries and veins (see Fig 1b), causing the vessels to intertwine (i.e., to make a bundle). Aneurysms and AVMs are often found together in the same blood vessel system.
To prevent bleeding, embolization (a procedure to insert glue and coils into the aneurysm/AVM using a catheter to prevent it from bursting) is performed. To aid the embolization procedure, image segmentation (automatic detection of blood vessels in medical images) and skeletonization (extracting centerlines of blood vessels) are used to separate regions of AVMs and guide the catheter through the blood vessel system:
The problem is that these methods are not able to segment (“locate”) the small regions of aneurysms and AVMs, which are equally important in surgical practice as the large anomalies.
The goal of this thesis is to automatically detect blood vessel anomalies, such as aneurysms, AVMs, and vessel narrowings (Fig. 1c) using deep neural networks. Since the ground truth data on vessel abnormalities are scarce, the project would aim to design a simulation model of aneurysms and AVMs and use the model to train a deep neural network for detection and segmentation of vessel abnormalities. The validation of the results will be done by the clinical opinion of physicians. Based on the clinical opinions the models will need to be adjusted accordingly and the deep networks will need to be re-trained.
The student may choose the preferred programming language (Python and C++ are recommended). For visualization, VTK may be used.
Fig.2 Generated mesh models of blood vessels with aneurysms (white arrows)