Enhancing Covid-19 Detection in X-Ray Images Through Image Preprocessing and Transfer Learning

In this study, we assess the efficacy of convolutional neural networks (CNNs) for classifying chest X-ray images into COVID-19, Lung Opacity, Viral Pneumonia, and Normal categories using a dataset of 21,165 images from Kaggle. Through rigorous preprocessing, including RGB to grayscale conversion and contrast enhancement via contrast-limited adaptive histogram equalization (CLAHE), we aim to improve diagnostic accuracy. We explore four CNN architectures: VGG-16, ResNet-50, GoogleNet, and AlexNet, employing Adam and Nadam optimizers to determine the optimal model and optimization strategy for disease classification. This research highlights the importance of image preprocessing and careful selection of CNN architectures and optimizers in enhancing diagnostic precision. The study's outcomes not only provide insights into leveraging deep learning for medical imaging analysis, especially in detecting COVID-19 from X-ray images, but also lay a foundation for future research to explore further image preprocessing techniques and their impact on diagnostic accuracy. This research contributes to the broader effort of applying advanced computational methods to improve healthcare outcomes. It offers a promising approach to enhancing the accuracy and efficiency of medical diagnostics through technology. Our study reveals that applying a grayscale preprocessing technique significantly enhances the accuracy of models in diagnosing diseases. Specifically, GoogleNet and VGG-16 models were found to be highly effective in disease classification. When these models were fine-tuned using the Nadam optimizer, GoogleNet achieved an outstanding accuracy rate of 91% and VGG-16 achieved 90% of accuracy. This indicates the potential of combining advanced neural networks with optimized preprocessing for more accurate medical imaging analysis.