Adaptive Contrast Enhancement for Digital Radiographic Images using Image-to-Image Translation

Abstract Digital radiography in medicine is a widely used imaging method for obtaining visual information about the inside of a body. To prepare the acquired raw image for diagnostic evaluation, the contrast must be adjusted depending on the examined part of the body and the reason of acquisition. The contrast enhancement of an image can be considered as a style transfer or an image-to-image translation which is an important field in deep learning. Based on common methods like the pix2pix network that only translate from one domain into one other, we propose a method (cc-pix2pix) for translating into multiple domains in one training. We provide additional information about the examination to the network for a specific contrast adjustment. Compared to the pix2pix network, the ccpix2pix reduces the mean squared error by a factor of six and achieves an improvement of approximately seven percentage points based on the histogram intersection of source and target images.