The application of self-supervised learning tools in the analysis of medical images through the image context restoration approach

Machine learning, especially deep learning, has enhanced medical image analysis over the past years. Training a good deep learning model requires a large amount of labeled data. Therefore, increasing the performance of machine learning models using unlabeled as well as labeled data is an important but challenging problem. Self-monitored learning offers a possible solution to this problem. In this paper, in order to better exploit unlabeled images, we propose a novel self-supervised learning strategy based oncontext restoration. We validate the context restoration strategy with three common problems in medical imaging: classification, localization, and segmentation. To perform the classification, we apply and test it on the scanning screen of two-dimensional ultrasound images of the fetus; To localize abdominal organs, we apply and test it in CT images, and to divide brain tumors, we apply and test it in multimodal MR images. In all three cases, self-supervised learning based on context restoration learns useful semantic features and leads to improved machine learning models of the above tasks.

Keywords:

Self-supervised learning , medical image analysis , context reconstruction