Modality classification for medical images using multiple deep convolutional neural networks

Modality is a key facet in medical image retrieval, as physicians are likely interested in only one of e.g. radiology images, flowcharts, and pathology photos. Traditional methods of modality classification focus on features engineering and demand us to be familiar with the prior domain knowledge. Therefore, we introduce one novel deep convolutional neural network which can automatically learn features from data. Our experiments are performed on the medical image modality dataset of the public ImageCLEF 2013. After training multiple deep convolutional neural networks on a GPU, we further improve the performance by combining them. Result ranks favorably (74.90%, within the top three groups) and shows the potential of the deep learning architecture in this field. Copyright © 2015 Binary Information Press.