This website requires JavaScript.

Segmentation of Parotid Gland Tumors Using Multimodal MRI and Contrastive Learning

Zi'an XuYin DaiFayu LiuBoyuan WuWeibing ChenLifu Shi
Aug 2022
Parotid gland tumor is a common type of head and neck tumor. Segmentation ofthe parotid glands and tumors by MR images is important for the treatment ofparotid gland tumors. However, segmentation of the parotid glands isparticularly challenging due to their variable shape and low contrast withsurrounding structures. Recently deep learning has developed rapidly, which canhandle complex problems. However, most of the current deep learning methods forprocessing medical images are still based on supervised learning. Compared withnatural images, medical images are difficult to acquire and costly to label.Contrastive learning, as an unsupervised learning method, can more effectivelyutilize unlabeled medical images. In this paper, we used a Transformer-basedcontrastive learning method and innovatively trained the contrastive learningnetwork with transfer learning. Then, the output model was transferred to thedownstream parotid segmentation task, which improved the performance of theparotid segmentation model on the test set. The improved DSC was 89.60%, MPAwas 99.36%, MIoU was 85.11%, and HD was 2.98. All four metrics showedsignificant improvement compared to the results of using a supervised learningmodel as a pre-trained model for the parotid segmentation network. In addition,we found that the improvement of the segmentation network by the contrastivelearning model was mainly in the encoder part, so this paper also tried tobuild a contrastive learning network for the decoder part and discussed theproblems encountered in the process of building.