Federated Learning for Multi-Center Medical Image Classification Using Deep Learning Models
Noor Salah Hassan, Ali Hussein Hamad · Ingénierie des systèmes d information · 2025
Artificial intelligence is being applied in numerous industries, including healthcare among others.Due in great part to needs like dependable findings, data security, exact prediction, and a volume of data, among other things, research is being undertaken in the AI-enabled healthcare market.Regarding conventional deep learning models, datasets saved on a single device are used throughout the training process.Training the data calls for both highly efficient equipment and a lot of storage capacity.The work shown here suggests a federated learning approach suitable for five different customers.9702 ultrasonic images of the gallbladder (GB) correspond with eight distinct disease types.Every client owns a part of the dataset with some unique classes from those of other clients.This is so since clients have divided the dataset.Two deep learning models applied and assessed in this work were CNN and VGG16.Clients used both models as well as the global ones.This paper proposes a possible global model solution based on the FedAvg aggregation method.The results show that VGG16 shows better outcomes in classification for both the client and the global model with a 99% accuracy rate in FL and a 94% accuracy rate for local training alone operations.CNN shows accuracy with a 99% in Florida and an 81% for local training initiatives.