Chest radiographs are one of the most common diagnostic tools that have made the detection and diagnosis of many lung diseases possible for doctors. Most hospitals have Picture Archiving and Communication Systems, which can store huge amounts of radiological information, including hundreds of X-rays. These devices allow doctors to make quick diagnoses and, relatively easily, observe what happens in our bodies. However, there are still several challenges associated with the use of these robust and unbalanced data from hospital databases that contain crucial imaging information. Deep learning models require massive data. They process large amounts of data for reliable and accurate computer-aided diagnosis systems. This paper references the “NIH Chest X-ray” dataset, comprising 112,120 frontal-view X-ray images of 30,805 patients, labeled as having one of 14 diseases. Many labels per image are assigned from radiology reports through natural language processing. This helps physicians find common chest diseases and locate them within the image. For demonstration, we used a unique approach: a unified multi-label image classification and disease localization framework–we tested this with our dataset using various functionalities. We have developed a deep learning model for the classification of thoracic disorders and also for detecting pathology using the pre-trained models-Inception V3, MobileNet, DenseNet 121 and DenseNet 201. It performs at 98.62% accuracy, 0.97 specificity, and 0.47 sensitivity, then used ResNet to bound boxes in 14 diseases for the location of the pathology region. The efficiency of DenseNet 201 was higher than others, as it shows a significant AUC curve for cardiomegaly. Given its promising early results, DCNN remains challenging for reading chest radiographs, since high precision in fully automated CAD systems demands more than image-level labels. Thus, we conclude that our approach remains superior to state-of-the-art techniques through rigorous experiments performed on ChestX-ray14, and the effectiveness of our approach is validated both numerically and visually.