A Diagnostic System for Intracranial Saccular and Fusiform Aneurysms with Location Detection

Brain aneurysm detection is challenging due to multiple types of aneurysms, different image modalities, small size, and occurrence of aneurysms on multiple locations. Majority of brain aneurysm detection systems target only single class of aneurysm (i.e. saccular). Additionally, the detection of aneurysm location is largely unexplored. To overcome these challenges, we propose a robust two-stage brain aneurysm diagnostic system capable of detecting both forms of aneurysm (saccular and fusiform) along-with the detection of location. The first stage involves the aneurysm detection where vasculature extraction is performed initially using the morphological and enhancement operations. Next, histogram statistics are employed to obtain the potential aneurysmal region of interests (RoIs) that are later reduced using our proposed automated technique. We proposed a 28-D feature vector consisting of shape and texture features to represent these RoIs and used them to train a KNN classifier for aneurysm detection. The second stage focuses on the location detection where part of the vasculature with identified aneurysm is cropped and segmented via watershed segmentation. The distance of aneurysm to these segments is calculated and three smallest least distanced RoIs are identified. Next, we employed our 6-D shape features to represent these RoIs and fed to another KNN classifier for location detection. We achieved an accuracy of 95% for aneurysm detection and 82.6% for location detection on a dataset of 209 digital subtraction angiography (DSA) images acquired from the Henry Ford Hospital. These results signify the effectiveness of our system for aneurysm detection along-with its location identification.