Federated learning with integrated attention multiscale model for brain tumor segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• Brain tumors are a deadly condition requiring accurate identification and monitoring.
• Magnetic Resonance Imaging (MRI) is crucial for brain tumor diagnosis but raises privacy concerns due to sensitive patient data.
• Traditional centralized methods for brain tumor segmentation risk privacy violations.

Purpose of the Study:

• To develop a privacy-preserving federated learning approach for brain tumor segmentation.
• To introduce a novel model that optimizes segmentation performance while managing client resources.
• To enhance the accuracy and efficiency of brain tumor detection using AI.

Main Methods:

• A Reinforcement Learning-based Federated Averaging (RL-FedAvg) model was developed, integrating Federated Averaging (FedAvg) with Reinforcement Learning (RL).
• A Double Attention-based Multiscale Dense-U-Net (mixed-fed-UNet) model was employed, utilizing the RL-FedAvg algorithm for dynamic hyperparameter optimization.
• The model was trained and evaluated on the BraTs 2020 dataset, focusing on privacy preservation by keeping data decentralized.

Main Results:

• The proposed mixed-fed-UNet model achieved a high accuracy of 98.24%.
• The model demonstrated a strong performance with a Dice coefficient of 93.28% on the BraTs 2020 dataset.
• The RL-FedAvg approach effectively optimized the global model and managed client resources, outperforming existing methods.

Conclusions:

• Federated learning, particularly the RL-FedAvg approach, offers a viable and privacy-preserving solution for brain tumor segmentation.
• The mixed-fed-UNet model shows significant potential for improving the accuracy and efficiency of brain tumor detection in medical imaging.
• This privacy-preserving AI technique can advance collaborative model development in sensitive medical applications.