A machine leaning approach for abdominal aortic aneurysm severity assessment using geometric, biomechanical, and patient-specific historical clinical features

摘要

Recent studies monitoring severity of abdominal aortic aneurysm (AAA) suggested that reliance on only the maximumtransverse diameter (D_(max)) may be insufficient to predict AAA rupture risk. Moreover, geometric indices, biomechanicalparameters, material properties, and patient-specific historical data affect AAA morphology, indicating the need for anintegrative approach that incorporates all factors for more accurate estimation of AAA severity. We implemented amachine learning algorithm using 45 features extracted from 66 patients. The model was generated using the J48 decisiontree algorithm with the aim of maximizing model accuracy. Three different feature sets were used to assess the predictionrate: ⅰ) using D_(max) as a single-feature set, ⅱ) using a set of all features, and, lastly ⅲ) using a feature set selected via theBestFirst feature selection algorithm. Our results indicate that BestFirst feature selection yielded the highest predictionaccuracy. These results indicate that a combination of several specific parameters that comprehensively capture AAAbehavior may enable a suitable assessment of AAA severity, suggesting the potential benefit of machine learning for thisapplication.