InceptionV4 and SEResNet101: precise predictors of intracranial hemorrhage and collateral circulation post-ischemic...

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Ischemic stroke (IS) poses a significant global health challenge.
• Predicting intracranial hemorrhage (ICH) and assessing collateral circulation are crucial for IS patient outcomes.
• Current diagnostic methods for ICH and collateral circulation are limited.

Purpose of the Study:

• To develop an accurate prediction method for ICH and collateral circulation using deep learning (DL) models.
• To improve prognostic assessment for patients undergoing interventional treatment for IS.

Main Methods:

• A meta-analysis of relevant literature was performed.
• Five DL models (DenseNet169, InceptionResNetV2, InceptionV4, MobileNetV3Small, SEResNet101) were trained and tested on preoperative CT images.
• An MCAO mouse model was used to identify potential biomarkers.

Main Results:

• Artificial intelligence (AI) demonstrated high accuracy in predicting ICH from CT images.
• InceptionV4 and SEResNet101 models showed superior performance in diagnosing ICH and collateral circulation.
• Key biomarkers (Kdr, Lcn2, Pxn) for ICH and poor collateral circulation were identified.

Conclusions:

• The InceptionV4 or SEResNet101 algorithms combined with preoperative CT imaging offer accurate and rapid prediction of ICH and collateral circulation in IS patients.
• This study integrates radiomics and DL for an effective approach to managing IS patients.
• The findings support the use of advanced AI in clinical decision-making for ischemic stroke treatment.