TL;DR
This study introduces a novel predictive model for the successful occlusion of unruptured intracranial aneurysms (UIAs) after flow diverter (FD) treatment. By integrating both the physical shape of the aneurysm and the complex blood flow dynamics within it, the model offers a more accurate prediction than previous methods, paving the way for personalized patient management and treatment optimization.
What Problem Does This Paper Address?
Unruptured intracranial aneurysms (UIAs) are a significant health concern, and while flow diverters (FDs) have revolutionized their treatment by promoting gradual occlusion, some aneurysms remain unhealed. This incomplete occlusion poses a persistent risk of rupture and necessitates prolonged dual antiplatelet therapy (DAPT), which carries its own risks. Clinicians currently lack a precise tool to predict which aneurysms will successfully occlude and when, making it difficult to tailor postoperative management, including DAPT duration. Existing predictive models often focus on either morphological or hemodynamic factors in isolation, failing to capture the complex interplay between them. This paper aims to bridge this gap by developing a comprehensive model that leverages both types of data.
Methods
The researchers employed a multi-pronged approach to build and validate their predictive model.
- Data Collection and Preprocessing: They gathered data from 51 patients treated with stand-alone FDs across four neurovascular centers. This included pre-treatment 3D reconstructed aneurysm models and computational fluid dynamics (CFD) simulations to derive hemodynamic parameters.
- Feature Engineering: Key morphological parameters like the Ostium-to-Vessel Ratio (OsR) and Aneurysm-to-Artery Ratio (ArR) were calculated. Hemodynamic parameters included parent artery wall shear stress (WSS) and its change post-FD.
- Model Development: Several machine learning algorithms were tested, including logistic regression, Random Forest, XGBoost, and K-Nearest Neighbors.
- Model Evaluation: To ensure robustness, especially with a limited dataset, they used repeated 5-fold cross-validation, 0.632 bootstrap, and 0.632+ bootstrap methods.
- Explainability: Crucially, Shapley Additive Explanation (SHAP) analysis was used to understand the contribution of each feature to the model's predictions, providing insights into why a prediction is made.
- Tool Development: A web application was developed to allow clinicians to input patient-specific data and receive real-time occlusion probability predictions and SHAP force plots.
Datasets
The study utilized a retrospective dataset of 51 patients with single unruptured intracranial aneurysms treated with stand-alone FD between June 2018 and December 2022. This dataset was collected from four major cerebrovascular disease centers in China. The data included clinical information, pre-FD 3D aneurysm models, and CFD simulation results. No new benchmark datasets were introduced.
Key Insights and Ideas
- The Synergy of Morphology and Hemodynamics: The core idea is that neither the aneurysm's shape nor the blood flow alone is sufficient for accurate prediction. It's the combination that unlocks predictive power.
- The Criticality of Change: It's not just the absolute values of hemodynamic parameters that matter, but how they change after FD implantation, particularly the change in parent artery wall shear stress (WSS). This suggests that the FD's impact on the surrounding vasculature is a key indicator of healing.
- Explainable AI for Clinical Trust: The use of SHAP analysis is a significant step towards making AI models interpretable for clinicians. Understanding why a prediction is made builds trust and facilitates clinical adoption.
- A Practical Tool for Decision Support: The development of a web application translates complex research into a usable tool, offering real-time predictions that can directly inform patient management.
Key Results
- Model Performance: Logistic regression emerged as the top-performing algorithm, achieving an Area Under the Curve (AUC) above 0.75, indicating good predictive performance.
- Key Predictors: The five most influential variables for predicting UIA occlusion were: follow-up time (> 6 months), Aneurysm-to-Artery Ratio (ArR), Ostium-to-Vessel Ratio (OsR), parent artery wall shear stress (WSS), and the change in parent artery WSS.
- SHAP Insights: SHAP analysis revealed that the change of parent artery WSS was the most significant contributor to accurate and early prediction of aneurysm occlusion.
- Clinical Significance: The study identified that a shorter follow-up time was associated with non-occlusion, and a higher OsR was also linked to poorer outcomes, aligning with previous research.
Limitations and Future Work
- Small Sample Size: The authors acknowledge the limitation of a relatively small sample size (N=51) and emphasize the need for larger, multi-center retrospective studies to further optimize the model.
- Focus on Small Aneurysms: The majority of aneurysms included were small (<5mm), and the authors suggest extending the model's application to medium and large intracranial aneurysms.
- Generalizability: While the model shows promise, further validation across diverse patient populations and FD devices would enhance its generalizability.
- Future Directions: The researchers plan to expand the predictive model's application to other aneurysm sizes and aim for multicenter validation. They also highlight the potential for personalized follow-up schedules and DAPT regimen adjustments based on these predictions.
Summary updated