Covid19 Identification from Chest X-Ray Images using Local Binary Patterns with assorted Machine Learning Classifiers

Abstract

The novel corona virus caused by the SARS-CoV2 virus originated in Wuhan, China and spread globally. The massive outbreak of the virus resulted in millions of people being infected. Early detection of the virus is crucial in the complete recovery of the patient but can be fatal if detected in the later stages. The symptoms of the virus being similar to flu make it difficult to detect. This paper attempts an automated system for identification of the Covid19 virus infected images of chest X-Ray. The proposed method uses a dataset which has human chest X-Rays of non infected people as well as patients suffering from pneumonia and Covid19 virus infection. Local binary patterns with variations in its input parameters are used for feature extraction. The resulting feature sets are classified using several machine learning algorithms and ensembles of these individual models. Results of experimentation are obtained across 10 fold cross validation testing. Evaluation metrics accuracy, positive predictive value (PPV), sensitivity and f-measure are used to compare performance. Observations of the results show that the ensemble of RTree-RForest-KNN gives the best classification performance while ensemble models perform better than most individual classifiers. Comparing the input parameters of the LBP, the best performance is given by parameters R=6 (P=48) and R=7 (P=56) gives the best performance for the average of metrics for 10 fold cross validation in the proposed Covid19 identification method from chest X-Ray images.