Comparison of Ruptured Intracranial Aneurysms Identification Using Different Machine Learning Algorithms and...

Area of Science:

• Radiology
• Machine Learning
• Medical Imaging

Background:

• Intracranial aneurysms pose a significant health risk, with rupture leading to severe outcomes.
• Accurate prediction of aneurysm rupture is crucial for timely intervention and improved patient management.

Purpose of the Study:

• To develop and compare radiomics models using various machine learning algorithms for predicting ruptured intracranial aneurysms.
• To evaluate the differential performance of these models under identical data conditions.

Main Methods:

• Extracted 107 radiomics features from computed tomography angiography images of 576 patients.
• Constructed predictive models using 12 machine learning algorithms, including boosting methods.
• Validated model performance using area under the curve (AUC) in training and validation cohorts.

Main Results:

• Boosting algorithms (AdaBoost, Gradient Boosting, CatBoost) achieved the highest AUC values (0.889, 0.883, 0.864) in the validation cohort.
• These boosting models significantly outperformed other machine learning algorithms.
• AdaBoost demonstrated robust performance with a cross-validation AUC range of 0.842 to 0.918.

Conclusions:

• Radiomics models integrated with machine learning effectively predict ruptured intracranial aneurysms.
• Boosting algorithms show superior efficacy in this predictive task, highlighting their potential for clinical application.
• The choice of machine learning algorithm significantly impacts the prediction performance of radiomics models for aneurysm rupture.