Dilated Convolutional V-Net with Transformer Integration for Brain Tumor Segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuro-oncology

Background:

• Brain tumors necessitate early detection for effective treatment, but their complex nature complicates segmentation in MRI scans.
• Existing deep learning models struggle to capture essential channel-wise feature interdependence for accurate spatial and channel feature representation.
• Automated segmentation is vital for timely diagnosis and treatment planning in neuro-oncology.

Purpose of the Study:

• To propose Dilated Convolutional Transformer V-Net (DCTransVNet) for improved volumetric brain tumor segmentation in multimodal MRI scans.
• To enhance feature learning by effectively capturing spatial and channel-wise interdependencies.
• To improve segmentation performance using a novel Augmented Brain Tumor Segmentation Generative Adversarial Network (AuBTS-GAN).

Main Methods:

• Developed DCTransVNet, an encoder-decoder network integrating 3D dilated residual blocks and Efficient Paired-Attention blocks.
• Incorporated a Global Context module to capture long-range dependencies for comprehensive feature representation.
• Utilized AuBTS-GAN to generate synthetic MRI data for augmenting training datasets.

Main Results:

• DCTransVNet demonstrated superior performance in volumetric brain tumor segmentation compared to state-of-the-art methods.
• The model effectively captured multi-scale contextual features and channel-wise feature interdependence.
• Experiments on benchmark datasets (BraTS, MSD) validated the proposed method's efficacy.

Conclusions:

• DCTransVNet offers a significant advancement in automated brain tumor segmentation from multimodal MRI.
• The proposed attention mechanisms and generative adversarial network contribute to enhanced segmentation accuracy.
• This work provides a robust tool for clinical diagnosis and treatment planning in neuro-oncology.