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

Calderon-Ramirez, Saul; Shengxiang-Yang,; Moemeni, Armaghan; Elizondo, David; Colreavy-Donnelly, Simon; Chavarria-Estrada, Luis Fernando; Molina-Cabello, Miguel A.

The Corona Virus (COVID-19) is an internationalpandemic that has quickly propagated throughout the world. The application of deep learning for image classification of chest X-ray images of Covid-19 patients, could become a novel pre-diagnostic detection methodology. However, deep learning architectures require large labelled datasets. This is often a limitation when the subject of research is relatively new as in the case of the virus outbreak, where dealing with small labelled datasets is a challenge. Moreover, in the context of a new highly infectious disease, the datasets are also highly imbalanced,with few observations from positive cases of the new disease. In this work we evaluate the performance of the semi-supervised deep learning architecture known as MixMatch using a very limited number of labelled observations and highly imbalanced labelled dataset. We propose a simple approach for correcting data imbalance, re-weight each observationin the loss function, giving a higher weight to the observationscorresponding to the under-represented class. For unlabelled observations, we propose the usage of the pseudo and augmentedlabels calculated by MixMatch to choose the appropriate weight. The MixMatch method combined with the proposed pseudo-label based balance correction improved classification accuracy by up to 10%, with respect to the non balanced MixMatch algorithm, with statistical significance. We tested our proposed approach with several available datasets using 10, 15 and 20 labelledobservations. Additionally, a new dataset is included among thetested datasets, composed of chest X-ray images of Costa Rican adult patients