Semi-Supervised COVID-19 CT Segmentation Via Contrastive Learning Guided by CNN and Transformer

Abstract

At the end of 2019, a novel coronavirus spread around the world. The automatic detection of infected lung areas from CT images of patients with COVID-19 is essential to help doctors make accurate diagnoses and analyzes. The method based on deep learning can quickly segment the target region and has become a popular way. However, the success of most current medical images segmentation methods depend on a large number of labeled data, and performs poorly on training datasets lacking large-scale good annotation. Especially for medical images, segmentation labels usually require manual segmentation by professional doctors, which is difficult to obtain. Therefore, semi-supervised learning has become a promising solution. In order to effectively to use unlabeled data, this paper designs a method by using contrastive learning to guide CNN and Transformer to conduct consistent training. CNN and Transformer are respectively used to capture the local and global features of the image, and generate pseudo labels for cross supervision. In addition, in order to better mine the information on semantics. We will first calculate the entropy value in training processes, discard the pixels with higher entropy value in the prediction results, and then use contrastive learning loss to help CNN learn better representation, so as to achieve better segmentation effect. We conducted a lot of experiments on the COVID-19 Computed Tomography dataset to evaluate our methods. Compared with several existing advanced semi-supervised segmentation algorithms, our method achieves better segmentation results.

Note:
Funding Information: This work was supported by the Natural Science Foundation of Jiangsu Province under Grant BK20190079 and the National Key Research and Development Program of China under Grant 2022ZD0119900.

Declaration of Interests: The authors declare that they have no known competing fi nancial interests or personal relationships that could have appeared to influence the work reported in this paper.

Keywords: COVID-19 Computed Tomography, Contrastive Learning, Semi-supervised Learning, Medicine Image Segmentation, Transformer.