CTFC: A Convolution and Visual Transformer Based Classifier for Few-Shot Chest X-ray Images

Abstract

While there is only a limited number of samples available for a new disease, few-shot learning is usually adopted in the field of medical image analysis. In this paper, we propose a classifier named CTFC (Convolution and visual Transformer for Few-shot Chest X-ray images). The proposed classifier mainly includes two components, i.e., the feature extractor and the distance metric classifier. First, features are extracted from a small number of support set samples through the visual converter and ResNet50. The features are fused afterwards. Second, feature prototypes are calculated from these support set sample. Finally, the query set samples are classified according to the Euclidean distance. Experiments on open source datasets show that the proposed classifier has advantages in few-shot learning techniques for chest X-ray diagnosis.