Intracranial Vessel Segmentation in 3D High-Resolution T1 Black-Blood MRI.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neuroimaging

Background:

• Accurate segmentation of intracranial vasculature is crucial for diagnosing and monitoring cerebrovascular diseases.
• Non-contrast-enhanced black-blood MRI offers a safe alternative to contrast-enhanced techniques but presents challenges for vascular visualization.
• Convolutional neural networks (CNNs) have shown potential in medical image analysis, including segmentation tasks.

Purpose of the Study:

• To evaluate the feasibility of using CNNs for intracranial vascular segmentation on non-contrast-enhanced black-blood MRI.
• To assess the performance of the CNN model in terms of accuracy, artifact handling, and differentiation of vascular structures.

Main Methods:

• A dataset of 37 cases was used for training and validation.
• Convolutional neural networks were employed for automated segmentation of intracranial vessels.
• Qualitative and quantitative assessments were performed to evaluate the segmentation results.

Main Results:

• The CNN model demonstrated successful differentiation of intracranial arteries and veins.
• No significant degradation in segmentation quality was observed due to stent artifacts.
• The model showed comparable recognition of aneurysm recurrence to Time-of-Flight Magnetic Resonance Angiography (TOF-MRA).
• A Dice similarity coefficient of 0.72 was achieved, indicating promising quantitative performance.
• Some false-positive and false-negative results were noted.

Conclusions:

• CNN-based segmentation of intracranial vessels using non-contrast black-blood MRI is feasible.
• This approach shows potential for clinical application in neurovascular imaging, offering a non-contrast alternative.
• Further refinement may be needed to address false-positive and false-negative findings for improved clinical utility.