A transformer-based generative adversarial network for brain tumor segmentation.

Area of Science:

• Medical Image Analysis
• Artificial Intelligence in Medicine
• Neuroimaging

Background:

• Brain tumor segmentation is a complex task in medical imaging.
• Transformers offer advantages in capturing long-range dependencies, complementing Convolutional Neural Networks (CNNs).

Purpose of the Study:

• To develop a novel transformer-based generative adversarial network (GAN) for automated multi-modal brain tumor segmentation.
• To leverage the strengths of transformers and CNNs for improved segmentation accuracy.

Main Methods:

• Proposed a GAN architecture with a U-shaped generator incorporating transformer blocks and ResNet at the bottom layer.
• Employed deep supervision in the generator and a CNN-based discriminator with multi-scale L1 loss.
• Validated the method on BRATS2015, BRATS2018, and BRATS2020 datasets.

Main Results:

• Achieved comparable or superior performance to state-of-the-art methods on the BRATS2015 dataset.
• Demonstrated successful generalization capabilities on the BRATS2018 and BRATS2020 datasets.
• The proposed method effectively segments brain tumors from multi-modal MRI scans.

Conclusions:

• The novel transformer-based GAN is effective for automated brain tumor segmentation.
• The architecture shows promising results and generalization ability for clinical applications.
• This approach advances the field of medical image segmentation for neuro-oncology.