Prediction Model of Hemorrhage Transformation in Patient with Acute Ischemic Stroke Based on Multiparametric MRI...

Area of Science:

• Neurology
• Radiology
• Artificial Intelligence

Background:

• Intravenous thrombolysis is standard for acute ischemic stroke.
• Intracerebral hemorrhage transformation (HT) is a common complication.
• Predicting HT is essential for patient management.

Purpose of the Study:

• To develop a reliable pretreatment model for predicting HT.
• To evaluate the predictive value of radiomics features from various regions of interest (ROIs).
• To compare models using all ROIs, abnormal ROIs only, and combined clinical factors.

Main Methods:

• Extracted 5400 radiomics features from multiparametric MRI in 71 patients.
• Utilized LASSO for feature selection and Random Forest for model building.
• Validated models on an independent cohort, comparing different ROI combinations and clinical factors.

Main Results:

• The radiomics model using all ROIs achieved an AUC of 0.871 and accuracy of 0.848.
• This outperformed a model using only abnormal ROIs (AUC=0.831, accuracy=0.818).
• A combined model with clinical factors and radiomics yielded the best performance (AUC=0.911, accuracy=0.894).

Conclusions:

• Radiomics features from all ROIs, including normal ones, significantly aid HT prediction.
• The combined radiomics and clinical model offers superior accuracy for pretreatment HT prediction.
• This model can assist clinicians in diagnosing and managing stroke patients.