Pneumonia Classification and Analysis in Chest X-ray by Means of Convolutional Neural Networks

In recent years, access to medical imaging studies has increased, improving the medical information available. Chest radiography allows detecting pathologies related to some serious diseases, becoming essential for diagnosing pneumonia, as well as detecting masses, injuries, infiltrations, pneumothoraxes, among others. Based on its performance for medical image classification, the present paper addresses the analysis and classification of chest X-rays showing pneumonia using convolutional neural networks, considering both transfer-learning and networks trained from scratch. Also, an approach based on self-organized maps is proposed in order to study the quality of generated characteristics. An accuracy of 0.965 ± 0.008 was achieved for normal vs. pneumonia classification with FNR of 4.2%, which is consistent with the state of the art.