Using Fractal Dimension to Predict the Risk of Intra-Cranial Aneurysm Rupture with Machine Learning - A Preprint

Q: How were the models evaluated statistically?

Each model was assessed using five iterations of 5-fold cross-validation, with AUC summarized by the median across iterations. Wilcoxon signed-rank tests compared AUC scores between model pairs.

Aug 3, 2026

Upload by Theodora

Intracranial aneurysms that rupture cause substantial morbidity and mortality. Traditional clinical risk assessment such as the PHASES score supports decision-making, but machine learning may improve predictive accuracy. This preprint compares Random Forest, XGBoost, Support Vector Machine, and Multi-Layer Perceptron models using clinical and radiographic features, reporting Random Forest as best overall. Fractal dimension (Minkowski dimension) emerges as the most influential feature across all models.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages7

Reading Time18 min

AccessFree to download

Abstract

1 Introduction

2 Methods

2.1 Dataset and Feature Extraction

2.2 Modeling

3 Statistical Analysis

4 Results

USING FRACTAL DIMENSION TO PREDICT THE RISK OF INTRA-CRANIAL ANEURYSM RUPTURE WITH MACHINE LEARNING A PREPRINT Pradyumna Elavarthi 1 , Anca Ralescu 1 , Mark D. Johnson2 , and Charles J. Prestigiacomo2 1Department of Computer Science, University of Cincinnati 2Neurosurgery, College of Medicine, University of Cincinnati September 27, 2024 ABSTRACT Intracranial aneurysms (IAs) that rupture result in significant morbidity and mortality. While traditional risk models such as the PHASES score are useful in clinical decision-making, machine learning (ML) models offer the potential to provide more accuracy. In this study, we compared the performance of four different machine learning algorithms—Random Forest (RF), XGBoost (XGB), Support Vector Machine (SVM), and Multi-Layer Perceptron (MLP)—on clinical and radiographic features to predict rupture status of intracranial aneurysms. Among the models, RF achieved the highest accuracy (85%) with balanced precision and recall, while MLP had the lowest overall performance (accuracy of 63%) . Fractal dimension ranked as the most important feature for model performance across all models. 1 Introduction The rupture of intracranial aneurysms (IAs) and subsequent subarachnoid hemorrhage carry high rates of mortality and disability [1] . While only a minority of IAs rupture [2][3], predicting which aneurysms are at the greatest risk is crucial for informing clinical management. The relationship between IA size and rupture risk has long been established [2][4], but more recent studies suggest that size alone is insufficient for determining rupture risk, with small aneurysms ( <7 mm) comprising a significant proportion of ruptured IAs [5][6] . In an attempt to gain a more holistic representation of an aneurysm’s morphology, shape ratios and morphometric parameters have been investigated with variable results [7][8][9] . The PHASES score, which incorporates both clinical and radiographic factors, is widely used to assess IA rupture risk [3][10] . However, its performance can be limited, with significant weight being given to size and location within the model. With excitement in the medical community around artificial intelligence, machine learning techniques have been applied by several groups to correlate clinical and radiographic variables with rupture status. Many of these studies focus on regression-based models, which can lack performance in complex nonlinear relationships [11][12][13] . ML models, such as Random Forests (RF), Support Vector Machines (SVM), and neural networks, have shown promise in handling complex nonlinear relationships [11][14] . In this study, we applied various ML techniques, including RF, XGBoost (XGB), SVM, and Multi-Layer Perceptron (MLP), to a dataset of clinical and radiographic variables to assess their ability to predict IAs rupture status. We compared the performance of these models to determine which provides the best predictive power and evaluate how they compare to existing clinical scores like PHASES. 2 Methods 2.1 Dataset and Feature Extraction The dataset used in this study consisted of 178 samples with 58 features, excluding the rupture status. These features included clinical parameters such as age, sex, gender, location of the aneurysm, hypertension status, whether the patient has multiple aneurysms, and morphological features corresponding to the geometry of the aneurysm. The morphological features included sphericity, bifurcation or sidewall location, undulation index, sa:vol ratio, lacunarity, size (mm), dome height (mm), maximum Feret diameter, among others. Additionally, a novel feature, fractal dimension (Minkowski Dimension), was incorporated, as it has been shown to correlate with rupture status in a recent study [9] . 2.2 Modeling To model the data, we removed outliers that were more than two standard deviations away from the mean and imputed them with the median value of the corresponding feature. We also found th

FAQs

How were the models evaluated statistically?

Each model was assessed using five iterations of 5-fold cross-validation, with AUC summarized by the median across iterations. Wilcoxon signed-rank tests compared AUC scores between model pairs.

Aug 3, 2026

Upload by Theodora