Abstract TP3: Predicting Rupture Risk of Intracranial Aneurysms Through Radiomics Analysis

Abstract

Background: Intracranial aneurysm rupture is a devastating event. Accurate estimation of their rupture risk can help guide management decisions. However, existing prediction methods remain subjective and have limited accuracy. This study is aimed at using clinical and imaging data to create an accurate predictive clinical tool for identification of aneurysms at a higher risk of rupturing. Methods: A prospectively acquired database of patients scanned with 3T high-resolution MRI between 2018 to 2023 was analyzed. Symptomatic status was defined as aneurysms that caused a cranial nerve neuropathy due to compression, and/or presented with a sentinel headache. Saccular aneurysms were segmented within 3D Slicer using co-registered T1 and T1+Gd 3T images. Morphological and aneurysm wall enhancement (AWE) metrics were obtained from the 3D models. Radiomic features (RFs) were extracted using pyRadiomics extension of 3D Slicer. This database was utilized for training and testing different machine learning prediction models. Finally, a Naïve bayes model was then employed for constructing clinical nomograms (Figure). Results: A total of one hundred and four aneurysms were analyzed. Twenty-nine were symptomatic. There were 87 RFs significantly different between symptomatic and asymptomatic aneurysms. Four machine learning prediction models were created based on clinical, morphological, AWE and RFs data. The best predictive model was based on RFs, achieving an accuracy of 74%. Conclusion: RFs combined with clinical, morphological and AWE metrics could potentially be a valuable clinical tool to predict intracranial aneurysms rupture risk.