Building a Synthetic Vascular Model: Evaluation in an Intracranial Aneurysms Detection Scenario

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Intracranial aneurysms pose significant health risks and are often located on the Circle of Willis.
• Deep learning models demonstrate superior performance in detecting and monitoring these aneurysms.
• Accurate and extensive datasets are crucial for training effective deep learning models.

Purpose of the Study:

• To develop a comprehensive synthetic 3D model of the cerebral vascular tree.
• To generate a substantial dataset for training 3D convolutional neural networks for aneurysm detection.
• To evaluate the performance enhancement achieved through synthetic data augmentation.

Main Methods:

• A full synthetic 3D model was created to mimic cerebral arteries, bifurcations, and intracranial aneurysms.
• The model simulates vasculature geometry using 3D Spline interpolation.
• It replicates background noise characteristics from Time Of Flight Magnetic Resonance Angiography acquisitions.

Main Results:

• The synthetic model successfully mimics complex vascular structures and aneurysm shapes.
• A neural network was developed and trained using the synthetic dataset for aneurysm segmentation and detection.
• In-depth evaluation demonstrated significant performance gains due to synthetic data augmentation.

Conclusions:

• The developed synthetic model provides a valuable tool for generating realistic brain vasculature data.
• This approach enhances the capabilities of deep learning models in detecting intracranial aneurysms.
• Synthetic data augmentation is a promising strategy to improve the accuracy and robustness of aneurysm detection systems.