eduzhai > Applied Sciences > Engineering >

Does Non-COVID19 Lung Lesion Help? Investigating Transferability in COVID-19 CT Image Segmentation

  • Save

... pages left unread,continue reading

Document pages: 11 pages

Abstract: Coronavirus disease 2019 (COVID-19) is a highly contagious virus spreadingall around the world. Deep learning has been adopted as an effective techniqueto aid COVID-19 detection and segmentation from computed tomography (CT)images. The major challenge lies in the inadequate public COVID-19 datasets.Recently, transfer learning has become a widely used technique that leveragesthe knowledge gained while solving one problem and applying it to a differentbut related problem. However, it remains unclear whether various non-COVID19lung lesions could contribute to segmenting COVID-19 infection areas and how tobetter conduct this transfer procedure. This paper provides a way to understandthe transferability of non-COVID19 lung lesions. Based on a publicly availableCOVID-19 CT dataset and three public non-COVID19 datasets, we evaluate fourtransfer learning methods using 3D U-Net as a standard encoder-decoder method.The results reveal the benefits of transferring knowledge from non-COVID19 lunglesions, and learning from multiple lung lesion datasets can extract moregeneral features, leading to accurate and robust pre-trained models. We furthershow the capability of the encoder to learn feature representations of lunglesions, which improves segmentation accuracy and facilitates trainingconvergence. In addition, our proposed Hybrid-encoder learning methodincorporates transferred lung lesion features from non-COVID19 datasetseffectively and achieves significant improvement. These findings promote newinsights into transfer learning for COVID-19 CT image segmentation, which canalso be further generalized to other medical tasks.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×