Modern evaluation in deep learning algorithms have the potential to exceedingly increase the scope of automating medical imaging. In this research, we design a method, based on the deep learning concept, for the detection and classification of 14 thoracic diseases through chest X-rays acquired from the NIH Chest X-ray Database. We employ transfer learning and use two pre-trained architectures for convolutional neural networks: ResNet-50 and DenseNet-121, both of which were trained on ImageNet. To help overcome the class imbalance problem associated with the dataset, we employed a weighted count loss function that improved the model’s responsiveness to frequent conditions. The results of the experiments show the possibilities for effective classification based on the developed classifiers. The study revealed the fact that DenseNet-121 is the best architecture for the diagnosis of advanced pulmonary diseases such as Emphysema and Emphysema in contrast, ResNet-50 achieved the best results for mildly advanced diseases in conclusion, these findings suggest an expected of deep learning algorithms in fully automating the process of thoracic imaging for disease discovery which would capitalize on timely diagnosis of the patient leading to better patient outcomes.