Deep Learning Based Pulmonary Chest X Ray Abnormalities Identification

Chest X-ray abnormalities represent a major challenge for medical diagnosis and rapid and accurate patient treatment. This study focuses on tuberculosis and uses image processing techniques to address the challenges of interpreting chest x-rays. The goal is to classify whether the individual has TB or Normal.The first step we collect data of chest X-ray data set that includes normal cases and TB lung abnormalities. To improve the X-ray images, Image processing techniques are used to improve clarity and quality and provide a basis for further research. Procedure practices pre-trained CNN architectures to excerpt structures from chest radiographs. The duplicate collection includes both normal and TB cases and is taken from various sources, such as the Shenzhen and Montgomery datasets. Data enrichment techniques are used to improve the model and its generalization in different scenarios. After transforming the images, training and validating datasets are created as part of the workflow. We implemented two models: Densenet169 and Resnet50, was trained to show which model had accurate results. To increase robustness and avoid redundancy, the training process includes important steps such as data augmentation, dropout, learning rate, batch correction, and reduction. We get the final result with Densenet most accurate when compared with Resnet.