{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/adversarial-deformation-regularization-for","title":"Adversarial Deformation Regularization for Training Image Registration Neural Networks","arxiv_id":"1805.10665","date":"2018-05-27","proceeding":null,"authors":["Yipeng Hu","Eli Gibson","Nooshin Ghavami","Ester Bonmati","Caroline M. Moore","Mark Emberton","Tom Vercauteren","J. Alison Noble","Dean C. Barratt"],"abstract":"We describe an adversarial learning approach to constrain convolutional\nneural network training for image registration, replacing heuristic smoothness\nmeasures of displacement fields often used in these tasks. Using\nminimally-invasive prostate cancer intervention as an example application, we\ndemonstrate the feasibility of utilizing biomechanical simulations to\nregularize a weakly-supervised anatomical-label-driven registration network for\naligning pre-procedural magnetic resonance (MR) and 3D intra-procedural\ntransrectal ultrasound (TRUS) images. A discriminator network is optimized to\ndistinguish the registration-predicted displacement fields from the motion data\nsimulated by finite element analysis. During training, the registration network\nsimultaneously aims to maximize similarity between anatomical labels that\ndrives image alignment and to minimize an adversarial generator loss that\nmeasures divergence between the predicted- and simulated deformation. The\nend-to-end trained network enables efficient and fully-automated registration\nthat only requires an MR and TRUS image pair as input, without anatomical\nlabels or simulated data during inference. 108 pairs of labelled MR and TRUS\nimages from 76 prostate cancer patients and 71,500 nonlinear finite-element\nsimulations from 143 different patients were used for this study. We show that,\nwith only gland segmentation as training labels, the proposed method can help\npredict physically plausible deformation without any other smoothness penalty.\nBased on cross-validation experiments using 834 pairs of independent validation\nlandmarks, the proposed adversarial-regularized registration achieved a target\nregistration error of 6.3 mm that is significantly lower than those from\nseveral other regularization methods.","url_abs":"http://arxiv.org/abs/1805.10665v1","url_pdf":"http://arxiv.org/pdf/1805.10665v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"adversarial-deformation-regularization-for","repo_url":"https://github.com/yipenghu/label-reg","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"image-registration","task_name":"Image Registration"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1805.10665","atlas_url":"https://app.syntology.ai/?focus=1805.10665","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}