META-learning-based retinal pathology classification from optical coherence tomography images

Meta-learning has been proposed with the goal of achieving general artificial intelligence, which makes deep learning models imitate advanced organisms, using prior knowledge to quickly adapt to new learning tasks with just a small number of samples. This ability is especially important for medical image analysis when training samples with pathologies are sometimes limited. To make full use of available medical image data and improve classification results, we propose to apply model-agnostic meta-learning (MAML) and MAML++ for pathology classification from optical coherence tomography (OCT) images. MAML trains a set of initialization parameters using training tasks, by which the model achieves fast convergence in new tasks with only a small amount of data. MAML++ is an improved version, which overcomes some shortcomings of MAML. Our model is pretrained on an OCT dataset with seven types of retinal pathologies, and then refined and tested on another dataset with three types of pathologies. According to the experimental results, the classification accuracies of MAML and MAML++ reached 90.60% and 95.60% respectively, which are higher than the traditional deep learning methods with pretraining.