Dual U-Net-Based Conditional Generative Adversarial Network for Blood Vessel Segmentation with Reduced Cerebral MR...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Accurate segmentation of brain vessels in Time-of-Flight Magnetic Resonance Angiography (TOF-MRA) is crucial for diagnosing neurological conditions.
• Current AI models often require extensive training data, posing a challenge for rare or specific medical imaging datasets.

Purpose of the Study:

• To develop an efficient AI model for brain blood vessel segmentation in TOF-MRA images.
• To reduce the dependency on large training datasets for achieving high segmentation performance.

Main Methods:

• Proposed a conditional generative adversarial network (UUr-cGAN) with a U-Net and residual U-Net generator architecture.
• Employed data augmentation and regularization techniques to mitigate the effects of limited training data and prevent overfitting.

Main Results:

• Achieved an average precision of 89.52% and a Dice score of 87.23% in cross-validated experiments for brain blood vessel segmentation.
• Demonstrated comparable performance to state-of-the-art methods while utilizing substantially fewer training samples.

Conclusions:

• The UUr-cGAN model effectively extracts critical features from small datasets for TOF-MRA image segmentation.
• This approach offers a viable solution for brain blood vessel segmentation, outperforming other convolutional neural network (CNN)-based methods in scenarios with limited data.