Image Classification of Lung X-ray Images using Deep learning

Abstract

X-rays have been the best support for medical research to make better diagnoses that help in predicting the type of disease. Several machines capture X-ray images of different body parts like the Lungs, Teeth, hands, legs, etc. The role of X-ray images came up in medical research and became very important in diagnosing the health condition of a lung X-ray. In this paper, we propose a new pooling layer before sending the image into the dense neural network by considering the lung X-rays dataset where normal and pneumonia images are taken and using the convolutional neural network (CNN) we determine the condition of the X-ray and classify them into a Normal or Pneumonia. We evaluated our model using a confusion matrix, noted the metrics of precision and recall scores, and compared them with existing models. This paper explains the CNN algorithm deeply and tries to confirm that: (I) X-ray pictures of diseased lungs can be classified using deep learning techniques if the training data is substantial. (II) Adding the average pool layer at the end of the architecture can get better results than many standard existing models. (III) Hyperparameter tuning can improve the deep learning model accuracies and helps the model to perform better. (IV) With a proper amount of training, hyperparameter tweaking, and using data augmentation we can be able to get better accuracy than many existing CNN models with the lowest number of trainable parameters. This makes it possible to accurately automate the process of interpreting X-ray images that could avoid deep MRI and CT scans which may affect patients with high radioactive waves.