MRI Brain Tumor Classification based on Federated Deep Learning
Khanh Le Dinh Viet, Khiem Le Ha, Trung Nguyen Quoc, Vinh Truong Hoang
Abstract
The proliferation of artificial intelligence (AI) has the potential to revolutionize many industries, but its application is hindered by the shortage of large-scale data. Data in various domains often exist in isolated silos, necessitating privacy and security. In the meantime, the lack of access to medical privacy prevented the development of trustworthy systems for diagnosing deadly malignancies like brain tumors. In this study, we apply a federated learning algorithm known as Federated Averaging (FedAvg) to train a brain tumor classification system using decentralized data without requesting the exchange of sensitive data. The proposed framework’s hyperparameters are adjusted to enhance its effectiveness on both independently and identically (IID) and non-independently and identically distributed data (Non-IID). Additionally, we leverage four cutting-edge deep learning models, namely, VGG16, ResNet50, ConvNext, and MaxViT, to optimize classification accuracy. The proposed framework achieves a classification accuracy of 98.69% on IID data and over 93% on Non-IID data.