Automated Pneumonia Diagnosis using a Customized Sequential Convolutional Neural Network

An automated system that can detect pneumonia on chest X-rays is needed for the rapid diagnosis of the pathology. In the past, numerous attempts have been made to automate the task with varying degree of success. This paper presents a novel, 18-layer deep sequential convolutional neural network based model that is proven to outperform the state of the art system for this task. A publicly available pediatric chest X-ray images dataset consisting of 5,856 X-ray images have been exploited for the training and testing of the model. The model performs the 'normal' v. 'pneumonia' classification task with the classification accuracy of 0.9439 which is around 1.6% better than the state-of-the-art system. The model yields high sensitivity (i.e. 0.99) which is very promising. But the model's limitation is that it has yielded lower than expected specificity (i.e. 0.86). Further improvements may be possible by thoroughly investigating the application of techniques such as transfer learning, fine tuning and more relevant data augmentation.