What can we learn from machine learning studies on flow diverter aneurysm embolization? A systematic review

Abstract

BackgroundAs the use of flow diverters has expanded in recent years, predicting successful outcomes has become more challenging for certain aneurysms.ObjectiveTo provide neurointerventionalists with an understanding of the available machine learning algorithms for predicting the success of flow diverters in occluding aneurysms.MethodsThis study followed Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, and the four major medical databases (PubMed, Embase, Scopus, Web of Science) were screened. The study included original research articles that evaluated the predictive abilities of various machine learning algorithms for determining the success of flow diverters in achieving aneurysm occlusion.ResultsFive studies out of 217 were included based on our criteria. The included studies used various variables (patient demographics, aneurysm and parent artery characteristics, flow diverter and hemodynamic-related features, and angiographic parametric imaging) to predict flow diverter treatment outcomes. The machine learning algorithms used, along with their respective accuracy rates, were as follows: logistic regression (61% and 85%), support vector machine (88%), Gaussian support vector machine (90%), linear support vector machine (85%), decision tree (80%), random forest (87%), k-nearest neighbors (83% and 85%), XGBoost (87%), CatBoost (86%), deep neural networks (77.9%), and recurrent neural networks (74%).Two studies trained the machine learning models with both all features and the most significant features. Both studies observed that the accuracy of machine learning models decreased by removing the insignificant features.ConclusionThe current literature indicates that machine learning algorithms can be trained to predict the success of flow diverters with an accuracy of up to 90%.