Utilizing counterfactual image generation improves robustness and performance in contrastive learning for medical imaging.
Reducing redundancy in CT image datasets for contrastive pre-training significantly improves the performance of deep learning models on downstream classification tasks.