Superresolution recurrent convolutional neural networks for learning with multi-resolution whole slide images

Abstract

A recurrent convolutional neural network is supervised machine learning way to process images that has both properties of convolutional and recurrent networks. We propose Convolutional Neural Network (CNN) based approach and its advanced recurrent version (RCNN) to solve the problem of enhancing the resolution of images obtained from a low magnification scanner, also known as the image super-resolution (SR) problem. The given class of scanner produces microscopic images relatively fast and storage efficiently. However, those scanners generate comparatively low quality images than images from complex and sophisticated scanners and do not have the necessary resolution for diagnostic or clinical researches, therefore low resolutions scanners are not in demand. The motivation of this study is to determine whether an image with low resolution could be enhanced by applying deep learning framework such that it would serve the same diagnostic purpose as a high resolution image from expensive scanners or microscopes. We presented novel network design and complex loss function. We validate these resolution improvements with computational analysis to show an enhanced image give the same quantitative results. In summary, our extensive experiments demonstrate that this method indeed produces images which are same quality to images from high resolution scanners. This approach opens up new application possibilities for using low-resolution scanners not only in terms of cost but also in access and speed of scanning for both research and possible clinical use.