Feature fusion using deep learning modules and fuzzy c-means for medical disease recognition

The Internet of Medical Things allows for much better learning models, often used in doctor decision support systems. Hence, an important element is the generalization of models by extending these systems to many hospitals. In this paper, we propose a federated learning solution with dual-CNN. It is a hybrid solution that creates a second dataset modified by a superpixel solution based on the fuzzy c-means. Then, the original image with the modified ones is used in training the CNN. Extending the operation on federated learning allows for additional parameter tuning in superpixel methods. Our proposition was tested on two datasets: x-ray and COVID-19 Melspectrograms. The obtained results show great potential in this hybridization and reach higher results than the existing solutions.