The effect of discrete cosine transform on COVID-19 differentiation from chest X-Ray images: A preliminary study

Dimililer, K.; Sekeroglu, B.

The rapid increase in the number of COVID-19 cases and the similarities in other pneumonia symptoms have led the efforts to make artificial intelligence-based diagnostic systems more effective. The use of chest radiography images in the diagnostic researches of COVID-19 is at the focus of artificial intelligence studies because of high-speed and worthwhile image acquisition. Image preprocessing methods aim to improve the classification accuracy of the models by providing more robust input images. Discrete Cosine Transform is an image compression technique and can be applied as a preprocessing method for classification problems. The reduction of the number of irrelevant pixels within the image during the compression process provides a more efficient and informative feature extraction from the images. This study presents the investigation of the Discrete Cosine Transform (DCT) and a deep convolutional neural network implementation on COVID-19 research using x-ray images. Two different datasets containing COVID-19 X-ray images were considered, and various experiments were performed by applying DCT on these images. The preliminary results of the two datasets showed that the implementation of DCT in the preprocessing phase of deep convolutional neural networks could increase the classification rates up to 10% and 5% in terms of μSensitivity and μROC AUC score, respectively. © 2021 IEEE.