Abstract
Background:
Ischemic stroke (IS) is a major global health issue. The risk of intracranial hemorrhage (ICH) after interventional treatment and the status of collateral circulation significantly affect patient prognosis. Traditional diagnostic methods for predicting ICH and collateral circulation are limited. This study aimed to develop a more accurate prediction method using deep learning (DL) models.
Methods:
A meta-analysis was conducted on relevant literature. Five DL models (DenseNet169, InceptionResNetV2, InceptionV4, MobileNetV3Small, and SEResNet101) were trained and tested with preoperative CT images from 58 patients and the CQ500 dataset. An MCAO mouse model was established to identify biomarkers.
Results:
AI showed high accuracy in predicting ICH from CT images. InceptionV4 and SEResNet101 outperformed other models in diagnosing ICH and collateral circulation. Kdr, Lcn2, and Pxn were identified as key biomarkers for ICH and poor collateral circulation.
Conclusion:
The InceptionV4 or SEResNet101 algorithm, when combined with preoperative CT imaging, enables accurate and rapid prediction of intracranial hemorrhage and collateral circulation following interventional treatment in patients with ischemic stroke. This study presents an effective approach that integrates evidence-based medicine, radiomics, and deep machine learning technologies.
Keywords:
collateral circulation; computed tomography; deep learning; intracranial hemorrhage; ischemic stroke.