Chest X-Ray Abnormality Detection Based on SqueezeNet

Abstract

Many Chest X-Rays are used by radiologists worldwide to identify the presence of chest diseases. Reading too many X-Rays in busy health centers may cause time and money loss. In addition, expert skill and concentration are required in the diagnosis of the disease. Errors or delays in the diagnosis of the disease can cause the patient to have worse ailments. In this study, to find solutions to these problems deep learning was used for chest X-ray disease detection. In this study, 660 chest X-Ray images taken from ChestX-Ray14, which has the largest database, were applied to SqueezeNet which is a convolutional neural network as a test data after being pre-processed, and classified as normal and abnormal. Transfer learning was used as a training style. The classification layer of the previously trained network with certain weights is adapted to 2 classes normally and abnormally. By changing the hyper parameters of the network, 90.95% success was achieved as a result of different trials. © 2020 IEEE.