Prediction of Cerebral Aneurysm Rupture Using Hemodynamic, Morphologic and Clinical Features: A Data Mining Approach

Cerebral aneurysms pose a major clinical threat and the current practice upon diagnosis is a complex, lengthy, and costly, multicriteria analysis, which to date is not fully understood. This paper reports the development of several classifiers predicting whether a given clinical case is likely to rupture taking into account available information of the patient and characteristics of the aneurysm. The dataset used included 157 cases, with 294 features each. The broad range of features include basic demographics and clinical information, morphological characteristics computed from the patient's medical images, as well as results gained from personalised blood flow simulations. In this premiere attempt the wealth of aneurysm-related information gained from multiple heterogeneous sources and complex simulation processes is used to systematically apply different data-mining algorithms and assess their predictive accuracy in this domain. The promising results show up to 95% classification accuracy. Moreover, the analysis also enables to confirm or reject risk factors commonly accepted or suspected in the domain.