Enhanced Lung Disease Detection through Transfer Learning Techniques

Lung abnormalities are prevalent among individuals of all age groups and may occur owing to multiple factors. The World Health Organization (WHO) reports that lung disorders account for millions of deaths worldwide annually. Pneumonia is a predominant cause of mortality, resulting in the deaths of nearly 2.5 million individuals. Lately, the lung infection caused by SARS-CoV-2 has impacted a broader worldwide population, prompting the World Health Organization (WHO) to classify it as a pandemic. Alleviating the effects of pulmonary illnesses is crucial for the general population, underscoring the need of inventiveness intended for management and prevention and to enhance health prospects worldwide. For medical image interpretation, deep learning has become a crucial asset for industrial and academic researchers pursuing exploration in this domain. This work inspects various image pre-processing practices, such as histogram equalization, total variation de-noising, Gaussian blur and CLAHE, utilized on a chest CT scan and X-ray images dataset. By improving image quality, these methods attempt to help the model classify diseases more accurately. The investigation looks into multiclass classification that includes COVID19, tuberculosis, viral pneumonia, bacterial pneumonia, and normal pneumonia. The results of using the deep learning models VGG16, VGG19, ResNet50, and DenseNet121 to identify diseases utilizing chest X-ray images are evaluated and contrasted. According to the results, VGG16 has the best accuracy, achieving 95.32% for training and 90.59% for testing.