A pilot study using a machine-learning approach of morphological and hemodynamic parameters for predicting aneurysms...

Area of Science:

• Cerebrovascular disease research
• Medical imaging analysis
• Computational fluid dynamics in medicine

Background:

• Identifying cerebral aneurysm rupture risk is crucial for clinical decision-making.
• Straightforward classification methods are needed to assess aneurysm instability.
• Existing methods often lack comprehensive integration of geometrical, hemodynamic, and clinical factors.

Purpose of the Study:

• To investigate the predictive importance of various risk factors for aneurysm wall enhancement.
• To evaluate the performance of multiple machine learning models in classifying aneurysm stability.
• To determine the relative significance of geometrical, hemodynamic, and clinical factors, including the PHASES score, using machine learning.

Main Methods:

• Applied nine distinct machine learning models to a dataset of 65 aneurysm cases.
• Utilized tenfold cross-validation with five repeats for model optimization, with AUC as the cost parameter.
• Validated models on a separate test set, assessing performance based on accuracy significantly higher than the non-information rate (NIR).

Main Results:

• Gradient boosting achieved the highest performance (AUC = 0.98), followed by generalized linear modeling (AUC = 0.80).
• The size ratio emerged as the dominant predictor, followed by the PHASES score and mean wall shear stress.
• Four models (random forests, generalized linear modeling, gradient boosting, linear discriminant analysis) showed significantly higher accuracy than NIR.

Conclusions:

• Machine learning models can effectively predict the relative importance of diverse factors in aneurysm wall enhancement.
• Size ratio, PHASES score, and mean wall shear stress are critical parameters for predicting wall enhancement in cerebral aneurysms.
• These findings support the clinical utility of ML-driven approaches for aneurysm risk stratification.