Deep Learning Algorithm Enables Cerebral Venous Thrombosis Detection With Routine Brain Magnetic Resonance Imaging.

Area of Science:

• Neurology
• Radiology
• Artificial Intelligence

Background:

• Cerebral venous thrombosis (CVT) is a rare but serious cerebrovascular condition.
• Diagnosis often relies on routine brain magnetic resonance imaging (MRI).
• Accurate and timely detection of CVT is crucial for patient outcomes.

Purpose of the Study:

• To develop and evaluate a novel deep learning (DL) algorithm for detecting CVT.
• To assess the algorithm's diagnostic performance using routine brain MRI.
• To compare the DL algorithm's accuracy against human radiologists.

Main Methods:

• Collected routine brain MRI scans (T1-weighted, T2-weighted, FLAIR) from patients suspected of CVT.
• Developed and trained DL algorithms on a dedicated dataset using 5-fold cross-validation.
• Evaluated the optimal DL algorithm against diagnoses made by four independent radiologists on a separate test set.

Main Results:

• The optimal DL algorithm achieved an AUC of 0.96, with 96% sensitivity and 88% specificity on a per-patient basis.
• On a per-segment basis, the DL algorithm showed 88% sensitivity and 80% specificity.
• The DL algorithm demonstrated significantly higher sensitivity than radiologists for both per-patient and per-segment diagnoses.

Conclusions:

• The developed DL algorithm significantly enhances the diagnostic performance for CVT detection using routine brain MRI.
• The algorithm exhibits high sensitivity and specificity, representing a promising tool for CVT diagnosis.
• This AI-driven approach offers a potential improvement in identifying cerebral venous thrombosis.