Development and validation of stability prediction models for intracranial aneurysms based on ensemble learning...

Area of Science:

• Neurosurgery and Medical Imaging
• Artificial Intelligence in Medicine
• Biomedical Data Science

Background:

• Intracranial aneurysms (IAs) pose a significant risk, and predicting their stability is crucial for patient management.
• Current methods for assessing IA stability have limitations, necessitating advanced predictive tools.

Purpose of the Study:

• To develop and validate ensemble learning models for predicting intracranial aneurysm (IA) stability.
• To evaluate the predictive performance of radiomics, conventional, and combined models using pre-rupture or pre-growth imaging features.

Main Methods:

• Retrospective collection of IA data from 7 hospitals, including computed tomography angiography (CTA), magnetic resonance angiography (MRA), and digital subtraction angiography (DSA).
• Extraction of radiomics, morphological, and clinical features from pre-rupture/pre-growth images.
• Development and evaluation of ensemble learning models (radiomics, conventional, combined) using univariate, multivariate, and recursive feature elimination (RFE) for feature selection.

Main Results:

• An internal dataset of 840 aneurysms and an external validation set of 271 aneurysms were analyzed.
• The radiomics model achieved an Area Under the Curve (AUC) of 0.85 in the external validation set.
• The conventional and combined models showed lower AUCs of 0.61 and 0.78, respectively, in external validation.

Conclusions:

• Radiomics features demonstrate significant potential for predicting the stability of unruptured intracranial aneurysms (UIAs).
• Integrating radiomics with conventional features did not improve the predictive power for IA stability.