eduzhai > Applied Sciences > Engineering >

4S-DT Self Supervised Super Sample Decomposition for Transfer learning with application to COVID-19 detection

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 10 pages

Abstract: Due to the high availability of large-scale annotated image datasets,knowledge transfer from pre-trained models showed outstanding performance inmedical image classification. However, building a robust image classificationmodel for datasets with data irregularity or imbalanced classes can be a verychallenging task, especially in the medical imaging domain. In this paper, wepropose a novel deep convolutional neural network, we called Self SupervisedSuper Sample Decomposition for Transfer learning (4S-DT) model. 4S-DTencourages a coarse-to-fine transfer learning from large-scale imagerecognition tasks to a specific chest X-ray image classification task using ageneric self-supervised sample decomposition approach. Our main contribution isa novel self-supervised learning mechanism guided by a super sampledecomposition of unlabelled chest X-ray images. 4S-DT helps in improving therobustness of knowledge transformation via a downstream learning strategy witha class-decomposition layer to simplify the local structure of the data. 4S-DTcan deal with any irregularities in the image dataset by investigating itsclass boundaries using a downstream class-decomposition mechanism. We used50,000 unlabelled chest X-ray images to achieve our coarse-to-fine transferlearning with an application to COVID-19 detection, as an exemplar. 4S-DT hasachieved a high accuracy of 99.8 (95 CI: 99.44 , 99.98 ) in the detection ofCOVID-19 cases on a large dataset and an accuracy of 97.54 (95 $ CI: 96.22 ,98.91 ) on an extended test set enriched by augmented images of a smalldataset, out of which all real COVID-19 cases were detected, which was thehighest accuracy obtained when compared to other methods.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...