Intracranial Aneurysm Rupture Risk Assessment Model Based on Multi-scale Deep Transfer Learning and Radiomics Features Fusion

Intracranial aneurysms are a severe health issue that can cause spontaneous subarachnoid hemorrhage (SAH) when they rupture, frequently leading to high mortality rates. Consequently, it is vital to assess the risk of aneurysm rupture accurately. In this study, we developed a comprehensive model for assessing the risk of aneurysm rupture. This model employs a combination of multi-scale deep transfer learning and radiomics feature fusion alongside the incorporation of essential clinical characteristics. Our research included 366 intracranial aneurysm cases, comprising 226 unruptured and 140 ruptured aneurysms. We utilized super-resolution reconstruction to enhance CT image quality and extracted and integrated deep features using the MedicalNet3D feature extractor from both original CT images and those reconstructed. We constructed three different classification models in our experiments and experimented with various feature fusion strategies. The findings revealed that the models performed best when multi-scale deep features, radiomics, and clinical characteristics were thoroughly integrated, with the SVM model showing extreme performance. In the training cohort, the SVM model achieved an AUC of 0.932 and an accuracy of 0.917; in the testing cohort, it achieved an AUC of 0.933 and an accuracy of 0.922. These results strongly support the benefit of using a multi-source data fusion strategy based on multi-scale deep learning features to improve the accuracy of predicting aneurysm rupture.