COVID-199 detection using deep learning on chest X-rays enhanced by adaptive methods

Deep Learning (DL) is the most popular Machine Learning (ML) algorithm today. Its use is becoming widespread day by day thanks to its high performance in classification in the field of Medical Image processing. Deep Learning is inspired by a neural network structure and includes many parameters. Due to its high performance, it is used in the classification of many diseases. Deep Learning algorithms, which are frequently used in the field of image processing, classify the pixels on the image by processing them in different layers. Before classifying the image with DL, it can be ensured that the classification is more successful with different preprocessing methods. In this study, the effect of different Deep Learning architectures on COVID was investigated using Local Histogram Equalization (LHE) and Adaptive Histogram Equalization (UHE), images were examined using Histogram Equalization and without using Histogram Equalization, and the effect of this filter on achievement was determined. The most successful parameters were determined by testing. Chest X-Ray images were used for classification, the dataset consists of COVID, Pneumonia and Normal images, Lung images were made in two separate arrangements on the dataset, and the first dataset was cropped as right and left lobe lung images, no clipping was performed on the second dataset. Each Histogram Equalization was tested with different parameters and the effect of each parameter on achievement was observed. Classification performance has been determined using different DL architectures such as DarkNet19, VGG16, AlexNet, MobileNet and Inception. These models were trained with a small dataset and tested on a larger dataset with learning transfer, so performance results were obtained by testing the weights learned with a small dataset on a large dataset.