Effectiveness of CAD using Preprocessed and Augmented Images for Detecting Lung Disease

Deep Learning has magnified growth in the medical image field in recent years. In the case of lung diseases, deep learning algorithms can automatically detect and classify abnormalities such as nodules, masses, infiltrates, and opacities on medical images with high accuracy. It is crucial to preprocess medical imagery to prepare the data for accurate and reliable analysis. Deep learning models can generate augmented versions of images to increase dataset diversity and improve model generalization. Variations of input images, such as rotations, flips, translations, or changes in lighting conditions, enrich the training data for better model performance. This paper demonstrates the impact of variations on training and validation accuracy of the CNN model, contributing to better performance of Computer-Aided Diagnosis(CAD) in detecting lung disease. The results show fewer augmentation leads to overfitting resulting in reduced validation accuracy. Strategic augmentation practices can effectively enhance training and validation accuracy resulting in an efficient CAD.