Automatic segmentation of hemispheric CSF on MRI using deep learning: Quantifying cerebral edema following large...

Area of Science:

• Neurology
• Radiology
• Artificial Intelligence

Background:

• Cerebral edema (CED) is a severe complication of acute ischemic stroke (AIS), particularly in large hemispheric infarction (LHI).
• Accurate assessment of CED is crucial for predicting patient outcomes.

Purpose of the Study:

• To implement a deep learning approach for extracting cerebrospinal fluid (CSF) from T2-Weighted Imaging (T2WI).
• To evaluate the relationship between quantified CSF volume and clinical outcomes in LHI patients.

Main Methods:

• Deep learning algorithm used to segment CSF from T2WI in 93 LHI patients.
• Hemispheric CSF ratio calculated to assess brain edema and prognosis.
• Receiver operating characteristic and multivariate logistic regression analyses performed.

Main Results:

• Automated CSF extraction achieved a mean Dice similarity coefficient of 0.830.
• Hemispheric CSF ratio accurately identified severe cerebral edema (AUC=0.867).
• Low hemispheric CSF ratio, previous stroke, and ASPECT score ≤6 predicted unfavorable outcomes.

Conclusions:

• Automated CSF volume assessment offers an objective biomarker for cerebral edema.
• Quantified CSF effectively predicts outcomes in LHI patients.
• This method aids in quantifying CED and confirming its prognostic value.