Classification of Covid-19 X-ray Images Using Tridiagonal Matrix Enhanced Multivariance Products Representation (TMEMPR)

Abstract

Medical images are crucial data sources for diseases that can not be diagnosed easily. X-rays, one of the medical images, have high resolution. Processing high-resolution images leads to a few problems such as difficulties in data storage, computational load, and the time required to process high-dimensional data. It is vital to be able to diagnose diseases fast and accurately. In this study, a data set consisting of lung X-rays of patients with and without COVID-19 symptoms was taken into consideration. Disease diagnosis from these images can be summarized in two steps as preprocessing and classification. The preprocessing step covers the feature extraction process and for this the recently developed decomposition-based method, Tridiagonal Matrix Enhanced Multivariance Products Representation (TMEMPR), is proposed as a feature extraction method. The classification of images is the second step where the methods of Random Forests and Support Vector Machines are applied. Also, the X-ray images have been reduced by 99,9% with TMEMPR and with several state-of-the-art feature extraction methods such as Discrete Wavelet Transform (DWT), Discrete Cosine Transform (DCT). The results are examined with regard to different feature extraction methods and it is observed that a higher accuracy rate is achieved when the TMEMPR method is used.