POSTER
Shabnam Bonyadi, Jared Matzke, Sithmi Jayasundara, Elizabeth Spurlock, Brad Sutton, Suguna Pappu
Background:
Predicting whether an aneurysm warrants intervention remains a major clinical challenge. Clinical decision-making relies heavily on aneurysm size and clinician judgement, yet size alone has not proven to be a reliable predictor. Recently, artificial intelligence (AI) methods have been increasingly applied to aneurysm imaging to capture complex features; however, limited work has synthesized findings across these diverse studies.
Methods:
This scoping review followed Joanna Briggs Institute methodology and PRISMA-ScR guidelines. A literature search was conducted on PubMed and Scopus using terms including Intracranial Aneurysm, Rupture, Prediction, Artificial Intelligence, Machine Learning, and Angiography. Two reviewers independently screened titles, abstracts, and full texts.
Results:
Rupture risk was associated with anterior circulation location. Morphological features, including aspect ratio, size ratio, bottleneck factor, non-sphericity, and irregularity, outperformed size alone. Radiomics features such as flatness, surface-to-volume ratio, and entropy-based metrics were predictive. Hemodynamic markers included low wall shear stress, high oscillatory shear index, increased relative residence time, and complex flow patterns. Models integrating clinical, morphological, radiomic, hemodynamic, and deep learning features demonstrated strong performance (AUC 0.85–0.95), with ensemble and boosting methods achieving accuracies >90%.
Conclusions:
AI applied to aneurysm imaging can extract meaningful features beyond size, improving rupture prediction. Future work should focus on developing integrative models that supports clinical decision making.