Optimizing Few-Shot Learning via Reptile Meta-Learning Approach for Toxoplasmosis Chorioretinitis Detection

Few-shot learning has emerged as a viable paradigm for addressing learning problems with minimal labeled instances, with recurrent issues such as overfitting and intricate feature extraction posing challenges that limit its effectiveness across various fields. In our study, we present a meta-learning technique that uses the Reptile algorithm to diagnose toxoplasmosis chorioretini-tis. Our approach capitalizes transfer learning, using cutting-edge Convolutional Neural Network (CNN) architectures-ResNet50, VGG-16, and Inception V3 as foundational models. Using the reptile meta-learning technique, we painstakingly test and compare these models' performance in diagnosing toxoplasmosis retinitis. This study represents an important step forward in the expanding field of meta-learning approaches for medical image classification as it adapts to tasks where the data is minimal, especially in the medical field where collection of large labeled data is difficult. It highlights Reptile's adaptability to the complexities inherent in ocular disease datasets, demonstrating its potential to significantly improve disease classification accuracy across a variety of medical imaging tasks. The proposed methodology emerges as a potent tool. In our evaluations, VGG-16 had an accuracy of 66%, ResNet50 had an accuracy of 80%, and Inception V3 had an accuracy of 84%. These findings not only support the usefulness of our technique but also demonstrate Inception V3's superior performance in distinguishing toxoplasmosis retinitis within the field of ocular illness detection.