# Artificial Intelligence in Cerebrovascular Imaging: A Targeted Review of Aneurysm Detection and Rupture Risk Prediction

## TL;DR

This review explores how artificial intelligence can improve detection and risk prediction for cerebral aneurysms, offering a more accurate and personalized approach to patient care.

## Contribution

The paper uniquely integrates radiomics, machine learning, and deep learning in a unified framework for aneurysm detection and rupture risk prediction.

## Key findings

- AI techniques like radiomics and deep learning reveal hidden patterns in imaging data for better aneurysm assessment.
- Machine learning models outperform traditional scoring tools in predicting aneurysm rupture risk.
- Hybrid AI models and multimodal data fusion are key for translating AI into real-world clinical settings.

## Abstract

Cerebral aneurysms are a potentially life-threatening vascular pathology that can lead to subarachnoid hemorrhage, a neurological emergency associated with high morbidity and mortality. Traditional imaging-based assessments (largely centered on aneurysm size, shape, and location) often fall short in accurately predicting rupture risk. This limitation highlights the need for more advanced, individualized diagnostic strategies. Recent advancements in artificial intelligence (AI) have introduced powerful tools capable of transforming cerebrovascular imaging and aneurysm management. This narrative review synthesizes published studies on the application of AI in cerebrovascular imaging, focusing on its potential to aid in aneurysm detection and rupture risk prediction. It examines the evolving role of AI through three primary technological approaches: radiomics, machine learning (ML), and deep learning (DL). Radiomics enables the extraction of quantitative features from imaging data, revealing patterns and morphological indicators that may not be visible to the human eye. ML models synthesize imaging, clinical, and hemodynamic data to predict rupture risk with greater precision than traditional scoring tools. DL techniques, particularly convolutional neural networks, automate aneurysm detection and interpretation directly from raw image data.

What sets this review apart from previous literature is its integrative approach: rather than focusing narrowly on one AI technique or imaging modality, it unifies radiomics, ML, and DL under a single framework and evaluates their clinical applications across both detection and risk prediction. Furthermore, it emphasizes emerging solutions like hybrid modeling, explainable AI, and multimodal data fusion, which are critical for real-world clinical translation. However, current AI-based methods remain at the investigational stage and have not yet been validated clinically, experimentally, or against existing diagnostic standards. Importantly, this review situates AI methods relative to established clinical benchmarks, including radiologist interpretation and risk scores such as PHASES (Population, Hypertension, Age, Size of aneurysm, Earlier subarachnoid, Hypertension, Age, Size of aneurysm, Earlier subarachnoid hemorrhage, and Site of aneurysm) and hemorrhage, and Site of aneurysm) and ELAPSS ( (Earlier subarachnoid hemorrhage Earlier subarachnoid hemorrhage, Location of aneurysm, Age, Population, Size of aneurysm, and Shape of aneurysm), Location of aneurysm, Age, Population, Size of aneurysm, and Shape of aneurysm), and emphasizes that rigorous prospective validation is essential before widespread adoption. It also proposes practical implementation strategies, including decision support integration, standardization protocols, and federated learning to enable secure data collaboration. By addressing both technical innovation and translational challenges, this review offers a clinician-focused roadmap that advances the field beyond theoretical models toward personalized aneurysm care. In doing so, it aims to reduce rupture rates and improve patient outcomes through precision medicine powered by AI.