Abstract
D
iabetic Retinopathy (DR) is the predominant and leading causes of blindness for people who have affected by diabetes in the world. DR Complication leads to affect the eyes and can lead to vision loss. Early detection and treatment are crucial for preventing or slowing the progression of the disease. In this study, we propose an approach for detecting diabetic retinopathy using feder- ated learning (FL).
A distributed machine learning technique called federated learning allows numerous devices to work together to jointly train a deep learning model with- out sharing their raw data. Each device in federated learning builds a local model on its own data, then aggregates the base model parameters to upgrade a global model. This process is repeated iteratively until convergence is reached. Computer-Aided Diagnosis frameworks are initially using machine learning and deep learning algorithms. DR diagnostic tools have been established in recent years using machine learning and deep learning models. these mod- els need big data for training and testing to validation of model behaviour. The Federated Learning utilizes the collaboration of multiple devices to train a deep learning model without compromising the privacy of individual patient data. Data dimensionality reduction and data cleaning and other exploratory data analysis process are carried as before implementing the model.
We show that federated learning can be used to overcome the problems caused by class imbalance when using real-world patient data. The main goal is to create a system that can control several medical facilities while maintaining data privacy. The findings indicate that the federated learning-based strat- egy is very accurate in identifying diabetic retinopathy and offers a potential technique for enhancing the early diagnosis and management of this condition. The proposed model outperforms existing state-of-the-art techniques in detect- ing DR and grading the severity of penetration levels while employing unseen fundus images, according to an analysis of observing performance metrics and model interpretation with reliability.
)