SDResU-Net: Separable and Dilated Residual U-Net for MRI Brain Tumor Segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuro-oncology

Background:

• Glioma is a common and aggressive primary brain tumor.
• Accurate tumor segmentation is crucial for diagnosis and treatment.
• Challenges include tumor heterogeneity and imaging noise.

Purpose of the Study:

• To develop a novel deep learning network for precise brain tumor segmentation.
• To improve the accuracy and efficiency of automated tumor segmentation in MRI scans.

Main Methods:

• Proposed SDResU-Net, a novel fully convolutional neural network (FCN) based on residual U-Net architecture.
• Integrated dilated and separable convolutions to enhance feature extraction and receptive field.
• Utilized separable convolution to process internal and inter-slice structures for better spatial information.

Main Results:

• SDResU-Net demonstrated superior performance compared to state-of-the-art methods on BraTS 2017 and BraTS 2018 datasets.
• The model effectively captured pixel-level details and spatial information for accurate segmentation.
• Cross-validation confirmed the model's robust generalization ability across datasets.

Conclusions:

• SDResU-Net offers a considerable alternative for automatic and accurate brain tumor segmentation.
• The proposed architecture enhances the capacity for local and global feature description.
• This method aids in improving diagnostic and therapeutic strategies for glioma patients.