An Unsupervised Parametric Mixture Model for Automatic Cerebrovascular Segmentation

Abstract

This chapter discusses one of the most critical neurological defects, cerebrovascular diseases and strokes, as they are a leading cause of many serious long-term disabilities. Developing accurate and fast methods for the early diagnosis and detection of potential stroke risk factors is crucial for preventing permanent damage, complications, and ultimately death. One of the most efficient ways for detecting stroke symptoms involves accurate segmentation of cerebrovascular trees and structures. Many modalities have been utilized for this purpose, such as ultrasound (US), magnetic resonance imaging (MRI), and computed tomography (CT). Due to many advantages over other modalities, we propose a novel segmentation method based on integration of statistical intensity models with the spatial interaction model for segmentation refinement. Further refinement is achieved by employing Gaussian scale space theory, followed by the majority voting schema and connectivity analysis for obtaining the final 3D segmentation of the cerebrovascular system.