Classification of Lung Diseases Using Transfer Learning with Chest X-Ray Images

Abstract

Millions of people worldwide are impacted by lung illnesses, which are a major cause of morbidity and mortality. An accurate and timely diagnosis is essential for managing and treating conditions effectively. A useful non-invasive technique for identifying and categorizing a range of lung disorders, chest X- ray imaging enables prompt treatment and better patient outcomes. This research describes an extensive study that develops and evaluates a classification based on deep learning model using the Chest X-ray images Dataset from National Institutes of Health (NIH), which covers a wide spectrum of thoracic anomalies. Astrong model could be developed thanks to the dataset, which included 112,120 images with annotations taken from radiological reports utilizing approaches from Natural Language Processing (NLP) techniques. Pre-trained architectures like ResNet and specially created Convolutional Neural Networks (CNNs), trained with a weighted loss technique to manage data imbalance, are among the models that have been put into practise. Through extensive exploratory data analysis, preparation procedures, model training, and evaluation, the research offers insights into the performance and accuracy of each illness class's classification. At 83.57%, ResNet50 has the highest accuracy, followed by Custom CNN at 78.25%. ResNet18's accuracy was 72.67%, whereas VGG19's accuracy was 67.12%. ResNet50 is the most accurate of these models, while VGG19 performs less accurately.