Comparison of Different Convolutional Neural Network Initialization Methods for COVID-19 Detection from X-Ray Images

COVID-19 is a pandemic that spread quickly to the whole world. Despite the development of various COVID-19 vaccines, the low vaccination rate in underdeveloped countries and the new COVID-19 variant are still threatening to prolong this pandemic. COVID-19 has severe effects on several body organs, especially on the lungs, resulting in features in the COVID-19 patients’ X-ray images distinctive from other viral cases of pneumonia. Although X-ray imaging is not a primary COVID-19 diagnosis method, it has been shown that deep learning methods, especially Convolutional Neural Networks (CNN), can successfully diagnose COVID-19 and detect COVID-19 induced lung abnormalities from X-ray images. This paper introduces a method to diagnose COVID-19 patients from chest X-ray images automatically. We present a method of adding random Gabor filters to the CNN layers to achieve higher classification accuracy in detecting COVID-19 from X-ray images and demonstrate the Gabor filters’ interpretable nature. The proposed method achieved 99% classification accuracy with a 95% confidence interval of (99.29%-99.33%) with std (±) 0.2874, 99% sensitivity, 99% specificity, 99% F1-score, and 96.2% Kappa.