Evaluating Pre-trained CNN Models for COVID-19 and Pneumonia Diagnosis Using Enhanced Medical Imaging

The COVID-19 pandemic has underscored the critical need for diagnostic methods that are both rapid and accurate. Chest X-ray (CXR) imaging is a valuable resource for detecting respiratory illnesses, including COVID-19, but its reliance on subjective visual analysis underscores the necessity for automated solutions. This research presents an innovative approach to image enhancement, designed to enhance the effectiveness of deep convolutional neural networks (CNNs) in tasks involving multiple class classifications. To evaluate its effectiveness, five pre-trained models—‘VGG16,’ ‘VGG19,’ ‘ResNet50,’ ‘ResNet101,’ and ‘EfficientNetB0’—have been tested on an enhanced dataset comprising three balanced categories: ’COVID-19’, ’normal’, and ’viral pneumonia’, each with 1,500 X-ray images. The proposed enhancement method addresses challenges such as low contrast and poor visibility, optimizing image quality for improved classification accuracy. Among the evaluated models, EfficientNetB0 demonstrated superior performance, achieving the highest ’precision’, ’accuracy’, ’F1-score’, and ’recall’. The findings highlight the effectiveness of integrating advanced image enhancement with pre-trained models to enable reliable and efficient COVID-19 diagnosis from CXR images.