Radiomic-based Risk Stratification of Intracranial Aneurysms
Andres Gudino1, Arshaq Saleem1, Sricharan Veeturi2, Diego Ojeda1, Elena Sagues Sese1, Sebastian Sanchez1, Carlos Dier1, Samantha Saenz1, Ariel Vargas1, Edgar Samaniego1
1Department of Neurology, University of Iowa Hospitals and Clinics, 2Canon Stroke and Vascular Research Center, University at Buffalo
Objective:

To determine if the radiomic analysis of intracranial aneurysms (IA) can stratify aneurysms between symptomatic and asymptomatic.

Background:

Radiomics is a quantitative imaging tool that can help in characterizing high resolution imaging at a voxel by voxel level, potentially enhancing the accuracy of IAs risk stratification.  We developed a radiomic-based normogram to characterize symptomatic aneurysms.  The normogram included relevant clinical information and radiomic features that were significantly different between symptomatic and asymptomatic aneurysms. 

Design/Methods:

High-resolution T3-MRIs of 90 patients with 104 IAs (75 asymptomatic, 29 symptomatic) were analyzed. Machine learning was used for training and testing subsets. The testing set was used to build four different nomograms, each incorporating different aneurysm wall enhancement analysis modalities, along with patients' demographic information. Finally, we evaluated the performance of each nomogram on an independent test set.

Results:

We identified 87 radiomic features that were significantly different between symptomatic and asymptomatic IAs. The best testing set was the RadScore nomogram that exhibited an accuracy of 77.1% (sensitivity: 89%, specificity: 73%, AUC: 0.83). 

Conclusions:

Radiomics derived from high-resolution MRI, combined with patient clinical data, can assist in identifying symptomatic IAs. This approach has the potential of triaging patients for treatment versus observation based on morphological characteristics.

10.1212/WNL.0000000000208182