Deformable image registration is often a slow process when using conventional methods. To speed up deformableregistration, there is growing interest in using convolutional neural networks. They are comparatively fast and canbe trained to estimate full-resolution deformation fields directly from pairs of images. Because deep learning-based registration methods often require rigid or affine pre-registration of the images, they do not performtrue end-to-end image registration. To address this, we propose a progressive training method for end-to-endimage registration with convolutional networks. The network is first trained to find large deformations at a lowresolution using a smaller part of the full architecture. The network is then gradually expanded during trainingby adding higher resolution layers that allow the network to learn more fine-grained deformations from higherresolution data. By starting at a lower resolution, the network is able to learn larger deformations more quicklyat the start of training, making pre-registration redundant. We apply this method to pulmonary CT data, anduse it to register inhalation to exhalation images. We train the network using the CREATIS pulmonary CT dataset, and apply the trained network to register the DIRLAB pulmonary CT data set. By computing the targetregistration error at corresponding landmarks we show that the error for end-to-end registration is significantlyreduced by using progressive training, while retaining sub-second registration times.
展开▼
