Purpose: Purpose of this study is the development and evaluation of an intelligent decision support system that will accurately predict the risk of rupture in incidentally detected intracranial aneurysms.
Patients-Methods: 100 patients with intracranial aneurysms (both ruptured and unruptured) detected by computed tomography angiography (CTA) and confirmed by invasive cerebral angiography (DSA) or clipping surgery were included. The intelligent decision support systems were developed based on measurements related to the geometry of the 100 aneurysms and their morphological characteristics, using conventional machine learning methods.
Results: The intelligent systems based on WEKAs Neural Network, WEKAs J48 decision tree and in Clips (a free software tool for building expert systems) had overall accuracy 80.6%, 80%, 73% respectively. These accuracy rates resulted from the true-positive classification of almost all ruptured aneurysms, as well as the false-positive classification of the most unruptured. This false-positive classification of unruptured aneurysms correlated with a specific morphologic feature (irregular shape).
Conclusion: The overall accuracy of the systems, deemed adequate. Their disability to correctly classify unruptured aneurysms reduces the overall success rates of all systems, however, this discrepancy associated to unruptured aneurysms of irregular shape, could well mean that such aneurysms are in higher risk of rupture and needs immediate treatment.