**4. Prediction of aneurysm complications**

The cornerstone of prediction modeling in aneurysm is to predict rupture, and statistical models such as logistic regression have been widely used for this purpose [35]. However, recent studies have demonstrated that machine learning (ML) models perform better than traditional statistical methods because they can process massive amounts of data and model nonlinear relationships [46].

Hemodynamics is considered the most valuable parameter in exploring intracranial aneurysm behavior. Promising AI tools, such as computational fluid dynamics, have been developed to assess hemodynamics [47]. Morphological features, including size and shape, have shown great potential in identifying aneurysms at risk of rupture, while geometric features that describe the 3D characteristics of the aneurysm can be automated to evaluate aneurysm formation, growth, and risk of rupture. Integrating clinical, morphological, and hemodynamic parameters can improve rupture prediction, but limited clinical use is still observed due to complexity, cost, and expertise requirements [5].

Several studies have used ML methods to predict complications arising from aneurysm rupture, such as vasospasm, delayed cerebral ischemia, and infarction [48]. Dumon et al. [49] developed an ANN prediction model that had a higher predictive value (AUC of 0.960) for symptomatic cerebral vasospasm than two multiple logistic regression models (AUC = 0.933 and 0.897). In another study, ML methods such as SVM, random forest, and multilayer perceptron outperformed logistic regression models in predicting delayed cerebral ischemia.

Tanioka et al. [47] used random forests to develop early prediction models for delayed cerebral ischemia, angiographic vasospasm, and cerebral infarction using clinical variables and matricellular proteins. The proteins osteopontin, periostin, and galectin-3 had prediction accuracies of 95.1%, 78.1%, and 3.8%, respectively. These studies demonstrate that ML methods have shown excellent performance in predicting complications that arise from aneurysm rupture.

Another application is the use of clinical data and CT perfusion from hospital admissions to predict outcomes of aneurysmal SAH. A random forest model was trained to predict dichotomized mRS (<2 and >2), and the accuracy was 84.4% in the training folds and 70.9% in the validation folds. However, it cannot be introduced into clinical practice because of small population size [50].
