Machine Learning Techniques for the Assessment of AAA Rupture Risk

Recent clinical studies have shown that the maximum transverse diameter of an abdominal aortic aneurysm (AAA) alone, or in combination with its expansion rate are not entirely reliable indicators of rupture potential. We hypothesize that AAA shape, size, and wall thickness may be related to rupture risk and can be deciding factors in the clinical management of the disease. A non-invasive, image-based evaluation of AAA size and geometry was implemented using an in-house code (AAAVASC v1.0, Carnegie Mellon University) on a retrospective study of 88 subjects. The contrast enhanced, computed tomography (CT) scans of 44 patients who suffered AAA rupture within 1 month of the scan were compared to those of 44 patients who received elective repair. The images were segmented and three-dimensional models were generated. Twenty-eight geometry-based indices were calculated to characterize the size and shape of each AAA and estimate regional variations in wall thickness. A multivariate analysis of variance was performed for all indices comparing the ruptured and non-ruptured data sets to determine which indices were statistically significant. A classification model was created using a J48 decision tree algorithm and its performance was assessed using 10-fold cross validation. The model correctly classified eighty-six data sets and had an average prediction accuracy of 74% (κ = 0.69). Such a decision model can be used in a clinical setting to assess the risk of AAA rupture with minimal user intervention.