Predicting Intracranial Aneurysm Rupture: A Multifactor Analysis Combining Radscore, Morphology, and PHASES...

Area of Science:

• Radiology
• Medical Imaging
• Computational Medicine

Background:

• Intracranial aneurysms (IAs) pose a significant rupture risk.
• Accurate prediction of IA rupture is crucial for patient management.
• Existing prediction methods may benefit from advanced computational approaches.

Purpose of the Study:

• To develop and validate a nomogram and machine learning (ML) models for predicting IA rupture.
• To integrate radiomics score (Radscore), morphology, and PHASES for enhanced prediction accuracy.
• To compare the performance of ML models against traditional logistic regression.

Main Methods:

• Retrospective analysis of 475 intracranial aneurysms (IAs) from 440 patients (2015-2023).
• Feature selection using t-tests and LASSO regression on radiomics data from digital subtraction angiography.
• Development of logistic regression (LR) models, a nomogram, and four ML algorithms (Random Forest, XGBoost, GBM, LightGBM).

Main Results:

• Multifactor LR models incorporating Radscore and PHASES (R+P) achieved an AUC of 0.899.
• Machine learning models demonstrated high performance, with AUCs ranging from 0.880 to 0.892 in the testing set.
• No significant performance difference was observed between ML models and the nomogram in the testing set (p > 0.05).

Conclusions:

• Nomogram and ML models integrating Radscore, morphology, and PHASES demonstrate high precision in predicting IA rupture.
• These models show excellent performance on calibration curves and decision curve analysis (DCA).
• The developed tools offer a promising approach for improving the prediction of intracranial aneurysm rupture.