Multi-modality brain tumor segmentation using dual-attention generative adversarial network.
Authors
Affiliations (5)
Affiliations (5)
- Department of Radiation Oncology, Graduate School of Biomedical Health Sciences, Hiroshima University, Hiroshima, Japan.
- Department of Radiation Oncology, MR Linac ART Division, Graduate School of Medicine, Chiba University, Chiba, Japan.
- Radiation Safety and Quality Assurance Division, National Cancer Center Hospital, Tokyo, Japan.
- Division of Medical Physics, Department of Information Technology and Medical Engineering, Human Health Sciences, Graduate School of Medicine, Kyoto University, Kyoto, Japan.
- Department of Carbon Ion Radiotherapy, Osaka University Graduate School of Medicine, Osaka, Japan.
Abstract
Brain malignant tumor needs accurate and fast segmentation for clinical diagnostics and radiation treatment planning. However, multi-contrast magnetic resonance images (MRI) are used to segment the necrotic, enhancing, and non-enhancing tumor (non-ET) regions. The current study improved an auto-segmentation model using a generative adversarial network (GAN) incorporated dual-attention (DAtGAN) on MRI images of glioma. A dataset used was that from the brain tumor segmentation (BraTS) challenge 2017 on Medical Image Computing and Computer Assisted Intervention Society (MICCAI). The DAtGAN introduced an attention module that helps to exploit the global information effectively for both the generator and discriminator of the GAN. The segmentation performance of the DAtGAN was compared with the conventional GAN. The average Dice similarity coefficient (DSC) of the GAN model was 0.85 for the ET, 0.89 for the core tumor region (CT), and 0.87 for the whole tumor (WT) region. The average DSC of the DAtGAN model was 0.88 for the ET, 0.92 for the CT, and 0.91 for the WT. The Intersection Over the Union was higher and the maximum Hausdorff distance (HD) was smaller for the DAtGAN than for the GAN model. The current study evaluated the auto-segmentation of glioma patients using the DAtGAN model. The proposed DAtGAN model is potentially valuable for improving the segmentation performance compared with other deep learning and atlas-based segmentation models. These results suggest that DAtGAN could be a valuable tool in clinical settings, providing more accurate and efficient segmentation for glioma patients, and ultimately enhancing treatment planning and outcomes.