eduzhai > Applied Sciences > Engineering >

Correcting Data Imbalance for Semi-Supervised Covid-19 Detection Using X-ray Chest Images

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 10 pages

Abstract: The Corona Virus (COVID-19) is an internationalpandemic that has quicklypropagated throughout the world. The application of deep learning for imageclassification of chest X-ray images of Covid-19 patients, could become a novelpre-diagnostic detection methodology. However, deep learning architecturesrequire large labelled datasets. This is often a limitation when the subject ofresearch is relatively new as in the case of the virus outbreak, where dealingwith small labelled datasets is a challenge. Moreover, in the context of a newhighly infectious disease, the datasets are also highly imbalanced,with fewobservations from positive cases of the new disease. In this work we evaluatethe performance of the semi-supervised deep learning architecture known asMixMatch using a very limited number of labelled observations and highlyimbalanced labelled dataset. We propose a simple approach for correcting dataimbalance, re-weight each observationin the loss function, giving a higherweight to the observationscorresponding to the under-represented class. Forunlabelled observations, we propose the usage of the pseudo and augmentedlabelscalculated by MixMatch to choose the appropriate weight. The MixMatch methodcombined with the proposed pseudo-label based balance correction improvedclassification accuracy by up to 10 , with respect to the non balanced MixMatchalgorithm, with statistical significance. We tested our proposed approach withseveral available datasets using 10, 15 and 20 labelledobservations.Additionally, a new dataset is included among thetested datasets, composed ofchest X-ray images of Costa Rican adult patients

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×