Enhanced brain tumour segmentation using a hybrid dual encoder-decoder model in federated learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Brain tumour segmentation is crucial for diagnostics and treatment planning.
• Conventional models face challenges in boundary delineation and generalization.
• Data privacy concerns hinder centralized training on multi-institutional datasets.

Purpose of the Study:

• To propose a Hybrid Dual Encoder-Decoder Segmentation Model for federated learning.
• To enhance segmentation accuracy, efficiency, and boundary delineation in brain tumour segmentation.
• To address data privacy limitations in large-scale medical imaging studies.

Main Methods:

• Integration of EfficientNet and Swin Transformer as dual encoders.
• Utilization of BASNet (Boundary-Aware Segmentation Network) and MaskFormer as decoders.
• Implementation within a federated learning framework to enable decentralized training.

Main Results:

• Achieved a Dice Coefficient of 0.94 and Intersection over Union (IoU) of 0.87.
• Demonstrated superior boundary delineation with Hausdorff Distance (HD95) of 1.61 and Boundary F1 Score (BF1) of 0.91.
• Reduced total training time through faster convergence in fewer federated rounds.

Conclusions:

• The hybrid model effectively enhances brain tumour segmentation accuracy and efficiency.
• Federated learning approach successfully preserves medical data privacy while achieving high performance.
• The integration of transformers, CNNs, and advanced decoders shows promise for future medical imaging segmentation tasks.