Radiomics Features in Contrast-Enhanced and Nonenhanced Magnetic Resonance Imaging Images Are Associated With High...

Aneurysm wall enhancement is a potential imaging biomarker for risk stratification of intracranial aneurysms (IAs). Variations in the texture of the magnetic resonance imaging (MRI) signal could shed light on the underlying pathobiology of the aneurysm wall. Radiomics can help quantify the textural complexity in MRI images, which could lead to better understanding and risk stratification of IAs. Herein, we investigated the potential use of radiomics derived from nonenhanced and contrast-enhanced MRI to identify high-risk IAs and evaluated their performance on different data sets.

We obtained 126 IAs from different centers and extracted radiomics features from nonenhanced and contrast-enhanced MRI for each aneurysm. We then built a random forest model from a part of the 3-T data set to identify high-risk IAs based on the 5-year population, hypertension, age, size of aneurysm, earlier SAH from another aneurysm, site of aneurysm (PHASES) score. We then tested the performance of this model on a part of the same 3-T data set, a 7-T data set, and an external 3-T data set. We also performed multivariate analysis to understand the significance of radiomics features.

We found that 75 radiomics features were significantly different between high- and low-risk IAs. The radiomics model had good performance when tested on the 3-T data set (accuracy, 90%; sensitivity, 86%; and specificity, 92%); however, when tested on external data sets, it had a moderate performance (accuracy, 88%; sensitivity, 50%; and specificity, 95% for external 3-T data set; and accuracy, 62%; sensitivity, 27%; and specificity, 100% for 7-T data set).

Radiomics derived from nonenhanced and contrast-enhanced MRI show high accuracy in identifying high-risk aneurysms from the same data set and could be used as a tool for quantifying aneurysm wall enhancement.