Semi-Supervised Medical Image Segmentation with Multimodal Data Augmentation and Self-Supervised Learning

Medical image segmentation is a critical task in clinical diagnosis and treatment. However, the high cost of acquiring annotated data and the scarcity of labeled datasets have made semi-supervised learning methods a focal point of research in this field. In this paper, we propose a semi-supervised medical image segmentation method that combines multimodal data augmentation, self-supervised pretraining, and pseudo-label generation and correction strategies. By leveraging complementary information from different modalities such as CT and MRI, along with self-supervised pretraining on unlabeled data, our model significantly improves segmentation performance under limited labeled data. Moreover, a confidence-based pseudo-label generation and correction strategy further enhances the label quality during the training process. Experimental results demonstrate that the proposed method outperforms traditional fully-supervised and semi-supervised approaches across multiple medical image segmentation datasets, with notable improvements in segmentation accuracy, especially in scenarios with scarce labeled data. This method provides an effective solution for medical image segmentation tasks and shows great potential for widespread application, particularly in medical image analysis with limited annotated data.