Over 2.2 billion people globally suffer from cataracts, which make early detection and treatment extremely difficult. The limitations of traditional diagnostic approaches arise from the absence of specialized knowledge and costly equipment. The absence of diverse and large-scale datasets impedes the progress of artificial intelligence (AI)-based cataract detection systems, which limits their accuracy to the clinical level and prioritizes patient data privacy while addressing dataset constraints by utilizing Federated Learning (FL) as a novel solution. Federated learning is a decentralized approach in machine learning that combines model updates either locally or at a central server, allowing multiple local devices to collaboratively develop a global model without the need to exchange raw data. This allows for model improvement across distant datasets while maintaining data privacy. We trained the Local Models (LM) simultaneously on multiple local servers using different datasets. Updated learning models (LMs) are combined using FL approaches to create a strong Global Model (GM), which improves detection accuracy and enables inter-institutional collaboration while preserving patient privacy. The paper focuses on putting this FL-based architecture into practice and evaluating it, providing a scalable and private cataract detection method. This research presents a scalable and privacy-preserving cataract detection method, leveraging FL to enhance diagnostic accuracy while addressing the global need for secure and effective tools in early cataract detection.