Cross-Domain Few-Shot Chest X-ray Recognition

Deep learning models in medical image analysis are often hindered by the lack of annotated images. To tackle this, Few-Shot Learning (FSL) has been introduced to improve performance with limited data. However, FSL models typically operate within a single data domain, which is a challenge since medical imaging involves diverse domains with significant variations due to different imaging techniques. In light of this, this paper proposes a data augmentation strategy and an alternating training manner to overcome domain discrepancies by simulating the X-ray imaging domain. Specifically, the data augmentation strategy is employed to simulate the image domain of X-ray imaging samples, enabling the model to adapt to the image domain through the cross-domain alternating training method. In the cross-domain classification process, the model’s generalization ability will not be reduced due to significant domain differences. To evaluate the performance of our model, we conducted experiments on the chest X-ray pneumonia dataset. The results demonstrate that under the 2-way 1-shot setting, our method achieves a significant improvement of 8.41% compared to the baseline model. Ablation and visualization also effectively explains the utility of the proposed method.