Medical image segmentation system for cerebral aneurysms

Background: Ruptured intracranial aneurysms is a much studied topic, with reports indicating the presence of unruptured intracranial aneurysms in approximately 5% of the adult population. Although the rupture rate of intracranial aneurysms is not high, it may lead to serious consequences including disability and mortality. Current treatments for intracranial aneurysms, however, also carry significant risks. To counter this, accurate assessment of the potential for in- tracranial aneurysm rupture is thereby essential in order for clinicians to balance the risk of surgery against the risk of the natural history. In current medical practice, greater emphasis is placed upon medical imaging technologies, including CTA, MRA and DSA scan for diagnostic purposes. These are widely applied in neurovascular imaging as a non-invasive diagnostic tool for the detection and evaluation of intracranial aneurysms. This makes it possible to visualize three dimensional (3D) cerebral aneurysms, the results allowing us to be able to reconstruct patient-specific vessels and aneurysms. Currently, the 3D geometry blood vessel has been applied in the performance of haemodynamic simulations, with the results obtained subsequently applied as a tool for the diagnosis of aneurysm risk and in support of neurosurgeons for the treatment of aneurysms. Visualization and haemodynamic simulations are all based upon the results of medical image reconstruction - aimed at the extraction geometries of targeted intracranial aneurysms from three-dimensional (3D) medical images. Despite the many image segmentation methods available, with varying approaches and algorithms, no dominant method yet exists, in terms of effectiveness, across the cerebral aneurysm. It has been indicated that it is necessary to develop a method in order to accurately segment the cerebrovascular aneurysm; thereby allowing us to measure aneurysm volume, size, and its 3D shape. Methods: In this thesis, I proposed a new method of segmentation called the Threshold-based Level Set (TLS) method. This method was specifically designed for application in cerebrovascular and cerebral aneurysms, and was based upon the Geodesic Active Contours model and Chan-Vese model (CV), integrating both region and boundary information to segment cerebral aneurysms through the use of a global threshold and gradient magnitude to form the speed function. Validation tests have been carried out to ensure the quality of the proposed TLS method in both 3D CTA scan and 3D DSA scan images. Both in-vivo and in-vitro validation tests were performed. In the in-vivo experiment, forty five aneurysm patients, including vascular and cerebral aneurysm CTA imagery across three locations; the internal carotid artery (ICA), middle cerebral artery (MCA) and anterior communicating artery (AComA), were used for the validation of the TLS method via analysis of the forty five TLS segmented models in terms of geometric shape, volume, and haemodynamic results. In the in-vitro experiment, however, a series of CT scans of silicone aneurysm models were conducted in this study, with four different silicone models and four rates of contrast agent dilution used to generate various image data sets for validation of the TLS method. Results: The proposed TLS method was found to be able to accurately segment intracranial aneurysms with blurred boundaries, complex cerebrovascular anatomical shapes and inhomogeneous images under automatic conditions. By comparison and contrast to other approaches, the TLS method revealed the highest volume overlap rate (JM), and lowest volume difference (VD), with its most important advantage featuring its ability to identify the complex local geometry of intracranial aneurysms - extremely important information in clinical application. The results from in-vivo validation showed that the TLS method appears higher in terms of overlap ratio and smaller in terms of volume difference than the other methods for ICA, MCA, and AComA aneurysms. The study likewise indicated that the volume differences and the overlap ratio of TLS may be controlled at a maximum of under 9% and a minimum of over 92% for all aneurysm locations respectively. In-vitro validation results showed that the TLS method was able to achieve over 89% of the volume overlap rate and under 7% of the volume difference across all different degrees of silicone model shape complexities and contrast agent dilutions. Conclusion: The TLS method is a technique with the ability to automatically segment intracranial aneurysms without the setting of a seed point or intensity threshold, and is likewise available for the segmentation of modifiable anatomical shapes, with blurred boundaries and inhomogeneous images. The TLS method may thus be a useful tool in the assistance of clinical diagnosis and surgical preparation, and play a vital part in the future of computational haemodynamics research.