Automated Chest X-ray Image Classification for Detecting Lung Diseases Using Deep Learning

The approach here was the design of an automatic lung disease classifier based on a deep learning strategy. The architectures specifically work in accordance with transfer learning techniques for improving the accuracy during classification, where we utilized the NIH Chest X-ray dataset and images from chest X-rays to develop our CNN architecture. We have trained our model in classifying 14 different thoracic diseases such as pneumonia, pneumothorax, and pleural effusion. Experiments could leverage data augmentation as well as other class balancing techniques to increase the ability of the model to generalize. Achieving an accuracy of 87.2%, the model achieved an F1-score of 0.80 and AUC-ROC of 0.92 on the test set. The results therefore imply deep learning techniques in detection of lung disease from chest X-rays would be practically useful as an adjunct tool for radiologists in medical diagnostics. The Grad-CAM interpretability technique was also applied to the model for more transparency in the decision-making process of the model. Also, by focusing on the most relevant regions of the chest X-ray, this work aims at reducing time while improving the accuracy of diagnosis, thus warranting better patient care outcomes.