Chest X-ray pathology detection using Deep Learning and Transfer Learning

Chest radiography is used to identify, diagnose, and treat lung illnesses including pulmonary nodules, TB, and interstitial lung disease. Chest radiography provides a wealth of information regarding a patient's condition. The doctor's ability to appropriately evaluate the data, on the other hand, is always a huge obstacle interpretation is significantly more challenging because of the chest X-overlaid ray's tissue features. It may be challenging to notice a lesion when there is little difference between it and the surrounding tissue, or when it spans the ribs or large pulmonary blood vessels. Neural network-based computer-aided detection systems are becoming essential and our study aims to develop deep learning models trained with Densenet, Mobilenet, VGG16, and Inception V3 models for the detection of 14-class pathologies in Chest X-Ray Ray images. The image sample dataset used for training consists of about 5606 images, each of which has a pathology that falls under one of 15 class labels and the evaluation of the model is done by finding the accuracy the of prediction of each class of 16 diseases by plotting confusion Matrix, ROC curves, and loss curves. Of all the models, we found that VGG-16 model is the least effective model for classification and best accuracy was achieved using Densenet followed by Mobilenet.