Effectiveness of Federated Learning and CNN Ensemble Architectures for Identifying Brain Tumors Using MRI Images

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Machine Learning

Background:

• Patient data privacy is a major concern in medical research, often leading to restricted access to sensitive information.
• Traditional machine learning requires centralized data, posing privacy risks and logistical challenges.
• Federated learning (FL) offers a solution by enabling collaborative model training without direct data sharing.

Purpose of the Study:

• To apply Federated Learning (FL) for privacy-preserving brain tumor classification using MRI images.
• To evaluate the performance of FL compared to traditional ensemble models in a medical context.
• To assess the scalability of the FL approach for larger datasets.

Main Methods:

• Trained multiple Convolutional Neural Network (CNN) models on MRI data.
• Selected the top three CNN models to create ensemble classifiers.
• Developed and trained a Federated Learning (FL) model using these ensemble architectures, processing model weights locally.

Main Results:

• The FL approach achieved 91.05% accuracy in brain tumor classification.
• A slight performance decrease was observed compared to the base ensemble model (96.68% accuracy).
• The method demonstrated scalability on a larger dataset, confirming its applicability.

Conclusions:

• Federated learning (FL) provides an effective method for privacy-protected brain tumor classification from MRI scans.
• FL maintains high accuracy while safeguarding patient data privacy.
• The FL approach is a viable and scalable alternative to traditional centralized deep learning methods in medical imaging.