Optimizing Brain Tumor Segmentation Networks Through Evolutionary Neural Architecture Search
Paper
| Paper/Presentation Title | Optimizing Brain Tumor Segmentation Networks Through Evolutionary Neural Architecture Search |
|---|---|
| Presentation Type | Paper |
| Authors | Yasmin, Farhana, Hasan, Mahade, Abdulla, Shahab, Hassan, Md Mehedi, Radjabovich, Radjabov Sukhro, Khader, Abdolraheem, Ahmed, Ali and Xue, Yu |
| Journal or Proceedings Title | Proceedings of 2025 International Conference on Digital Image Computing: Techniques and Applications (DICTA) |
| Number of Pages | 8 |
| Year | 2025 |
| Publisher | IEEE (Institute of Electrical and Electronics Engineers) |
| Place of Publication | IEEE |
| ISBN | 9798331571450 |
| 9798331571467 | |
| Digital Object Identifier (DOI) | https://doi.org/10.1109/DICTA68720.2025.11302470 |
| Web Address (URL) of Paper | https://ieeexplore.ieee.org/document/11302470 |
| Web Address (URL) of Conference Proceedings | https://ieeexplore.ieee.org/xpl/conhome/11302408/proceeding |
| Conference/Event | 2025 International Conference on Digital Image Computing: Techniques and Applications (DICTA) |
| Event Details | 2025 International Conference on Digital Image Computing: Techniques and Applications (DICTA) Parent International Conference on Digital Image Computing Techniques and Applications (DICTA) Delivery In person Event Date 03 to end of 05 Dec 2025 Event Location Adelaide, Australia |
| Abstract | Accurate brain tumor segmentation is essential for clinical diagnosis and treatment planning, yet manual design of optimal architectures is time-consuming and expertise-intensive. This paper proposes BrainEvoNAS, an evolutionary neural architecture search framework that automates UNet-based model design for brain tumor segmentation. BrainEvoNAS integrates low-and high-frequency feature extraction in the encoder, uncertainty-guided refinement in the middle, and multi-scale dilated convolution with skip fusion in the decoder. Using a comprehensive search space covering layers, filters, activations, pooling, kernels, dilations, and uncertainty weighting, it employs crossover, mutation, and fitness evaluation to evolve efficient architectures. Experiments on the BraTS 2021 dataset show BrainEvoNAS achieves a dice similarity coefficient (DSC) of 92.07 %, Hausdorff distance (HD95) of 1.58, with only 4.75 M parameters and 12.44 GFLOPs, outperforming state-of-the-art methods in both accuracy and computational efficiency. This framework provides a scalable, automated solution for accurate and efficient brain tumor segmentation, enabling practical clinical deployment. |
| Keywords | Brain tumor; Clinical deployment; Hausdorff distance; BrainEvoNAS |
| Article Publishing Charge (APC) Funding | Researcher |
| Contains Sensitive Content | Does not contain sensitive content |
| ANZSRC Field of Research 2020 | 429999. Other health sciences not elsewhere classified |
| Public Notes | Files associated with this item cannot be displayed due to copyright restrictions. |
| Byline Affiliations | Nanjing University of Information Science and Technology, China |
| UniSQ College | |
| University of South Australia | |
| King Abdulaziz University, Saudi Arabia |
https://research.usq.edu.au/item/101024/optimizing-brain-tumor-segmentation-networks-through-evolutionary-neural-architecture-search
24
total views4
total downloads14
views this month0
downloads this month