Artificial Intelligence (AI) has a number of significant applications across all verticals. The technology has special use cases in healthcare. Right from patient diagnosis to development of drugs, AI can play a major role. While this all this is true, training a neural network is the most daunting task in developing AI for healthcare.
The researchers from Nvidia and King’s College London are introducing a new method for training neural networks. The method called Federated Learning is a learning paradigm based on decentralized data. An algorithmic model is trained in multiple iterations at different sites using federated learning model. In healthcare, federated learning offers a degree of privacy for hospitals and other organisations. They can create their resource pool to train a deep learning model without actually sharing the data.
A degree of privacy for hospitals and other organisations is important in healthcare sector. The client-server federated approach designed by the researcher requires a centralized server to maintain a global deep neural network. With this approach, hospitals are given a copy of their network to train on their own dataset.
The training of dataset happens locally for a couple of iterations. The participants then send their updated version back to the server. The new consensus model is again shared wit hate participants where the training continues. Researchers used the brain tumour segmentation data from BraTS 2018 dataset for this research. The data included MRI scans of 285 patients. The model encourages collaborative training efforts, which is essential to have a clear impact on the advancement of healthcare AI.