Automated Cerebrovascular Segmentation and Visualization of Intracranial Time-of-Flight Magnetic Resonance...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Time-of-flight magnetic resonance angiography (TOF-MRA) is crucial for visualizing neurovasculature.
• Manual volume rendering (VR) for TOF-MRA is time-consuming and labor-intensive for radiologists.
• Deep learning (DL) offers potential for automating vessel segmentation in TOF-MRA.

Purpose of the Study:

• To evaluate the image quality of DL-based vessel segmentation for automated intracranial artery acquisition in TOF-MRA.
• To compare the performance of a proposed DL method against state-of-the-art techniques on external datasets.
• To assess the diagnostic and visualization capabilities of automated segmentation versus manual VR.

Main Methods:

• 394 TOF-MRA scans (cerebral vascular health, aneurysms, stenoses) were analyzed.
• A proposed convolutional neural network (CNN) and two other DL methods were evaluated for generalization.
• Experienced radiologists qualitatively assessed image quality (0-5 scale) and compared manual VR with CNN segmentation.

Main Results:

• The proposed CNN achieved superior clinical scoring on external datasets, with visualization comparable to manual reconstructions.
• High agreement (median score 5.0) was observed between CNN segmentation and manual VR for healthy intracranial arteries.
• Quantitative analysis showed high cerebrovascular overlap (Dice similarity coefficient 0.947 training, 0.927 validation).

Conclusions:

• Automated cerebrovascular segmentation using DL in TOF-MRA is feasible.
• The image quality of DL-based segmentation is comparable to expert manual VR in terms of vessel integrity, collateral circulation, and lesion morphology.
• DL-based segmentation provides adequate diagnostic quality and efficient workflow automation for neurovascular imaging.