Image Patch-Based Net Water Uptake and Radiomics Models Predict Malignant Cerebral Edema After Ischemic Stroke

Area of Science:

• Neurology
• Radiology
• Medical Imaging Analysis

Background:

• Malignant cerebral edema (MCE) following ischemic stroke is a critical determinant of patient outcomes, often necessitating surgical intervention.
• Accurate and early prediction of MCE is crucial for timely treatment decisions, such as decompressive craniectomy.
• Current methods for predicting MCE, like CT perfusion and lesion segmentation, are resource-intensive and require follow-up imaging.

Purpose of the Study:

• To introduce and validate a novel Image Patch-based Net Water Uptake (IP-NWU) method for predicting MCE.
• To assess the performance of IP-NWU using only non-enhanced admission CT scans, eliminating the need for lesion segmentation.
• To compare IP-NWU with existing Segmented Region-based NWU (SR-NWU) methods and develop radiomics models for enhanced MCE prediction.

Main Methods:

• Developed the IP-NWU procedure by comparing densities of ischemic and contralateral normal patches from non-enhanced admission CT scans.
• Compared IP-NWU with SR-NWU, which requires segmented lesion data from follow-up CT.
• Constructed radiomics predictive models for MCE by integrating IP-NWU and other imaging features using a random forest classifier.

Main Results:

• IP-NWU was significantly higher in patients with MCE compared to those without (p < 0.05).
• IP-NWU demonstrated strong predictive capability for MCE with an Area Under the Curve (AUC) of 0.86.
• No significant difference was observed between IP-NWU and SR-NWU in predicting MCE, with exceptional inter-reader and inter-operation agreement for IP-NWU (ICC = 0.92 and 0.95, respectively).
• The radiomics model combining IP-NWU with imaging features achieved the highest predictive performance with an AUC of 0.96.

Conclusions:

• IP-NWU is a reliable and efficient method for predicting MCE using readily available non-enhanced admission CT scans.
• The IP-NWU method offers comparable predictive accuracy to more complex segmentation-based approaches.
• Radiomics models incorporating IP-NWU show significant potential for precise MCE prediction, facilitating earlier clinical management of ischemic stroke patients.