TOWARD IMPROVED PREDICTION OF AAA RUPTURE RISK: IMPLEMENTATION OF FEATURE-BASED GEOMETRY QUANTIFICATION MEASURES COMPARED TO MAXIMUM DIAMETER ALONE

摘要

Data mining techniques are capable of extracting important relationships and correlations among large amounts of data while machine learning methodologies can utilize these correlations to generate models capable of classification and prediction. The combination of machine learning and data mining is an important contribution of the present work for two reasons: (1) given a large database of features that describe the geometry of native abdominal aortic aneurysms (AAAs), patterns and relationships in the data are derived that may not be apparent to the human eye, and (2) statistical models are generated that can classify new data and determine which features discriminate among different aneurysm populations. The objectives of this study were to use anatomically realistic AAA models to evaluate a proposed set of global geometric indices describing the size, shape and individual wall thickness of the aneurysm sac, and use a learning algorithm to develop a model that is capable of discriminating the rupture status of these aneurysms.