Additive Angular Margin for Few Shot Learning to Classify Clinical Endoscopy Images

Endoscopy is a widely used imaging modality to diagnose and treat diseases in gastrointestinal tract. However, varied modalities and use of different imaging protocols at various clinical centers impose significant challenges when generalising deep learning models. Moreover, the assembly of large datasets from different clinical centers can introduce a huge label biases in multi-center studies that renders any learnt model unusable. Additionally, when using new modality or presence of images with rare pattern abnormalities such as dysplasia; a bulk amount of similar image data and their corresponding labels may not be available for training these models. In this work, we propose to use a few-shot learning approach that requires less training data and can be used to predict class labels of test samples from an unseen dataset. We propose a novel additive angular margin metric in the framework of the prototypical network in few-shot learning setting. We compare our approach to the several established methods on a large cohort of multi-center, multi-organ, multi-disease, and multi-modal gastroendoscopy data. The proposed algorithm outperforms existing state-of-the-art methods.