Automatic classification of medical X-ray images with convolutional neural networks

Abstract

The classification of medical images is an important step for image-based clinical decision support systems. With the number of images taken per patient scan rapidly increasing, there is a need for automatic medical image classification systems that are accurate because manual classification and annotation is time-consuming and prone to errors. This paper focuses on automatic classification of X-ray image from the ImageCLEF 2009 dataset based on anatomical and biological information using the InceptionV3 model. The X-ray images are prepared and preprocessed with two different padding techniques, two image enhancement techniques and layering to convert the grey-scale images to 3-channel images to prepare them for InceptionV3. In terms of classification loss, constant padding with no enhancements had the best performance with an accuracy of 68.67% and a loss of 1.442. In terms of classification accuracy, constant padding with enhancement had the best performance with an accuracy of 71.34% and a loss of 1.608.