{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,1]],"date-time":"2026-10-01T06:16:55Z","timestamp":1790835415311,"version":"4.1.0"},"reference-count":44,"publisher":"Wiley","issue":"12","license":[{"start":{"date-parts":[[2023,10,17]],"date-time":"2023-10-17T00:00:00Z","timestamp":1697500800000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/onlinelibrary.wiley.com\/termsAndConditions#vor"}],"content-domain":{"domain":["aapm.onlinelibrary.wiley.com"],"crossmark-restriction":true},"short-container-title":["Medical Physics"],"published-print":{"date-parts":[[2023,12]]},"abstract":"<jats:title>Abstract<\/jats:title><jats:sec><jats:title>Background<\/jats:title><jats:p>Cone beam computed tomography (CBCT) plays an important role in many medical fields nowadays. Unfortunately, the potential of this imaging modality is hampered by lower image quality compared to the conventional CT, and producing accurate reconstructions remains challenging. A lot of recent research has been directed towards reconstruction methods relying on deep learning, which have shown great promise for various imaging modalities. However, practical application of deep learning to CBCT reconstruction is complicated by several issues, such as exceedingly high memory costs of deep learning methods when working with fully 3D data. Additionally, deep learning methods proposed in the literature are often trained and evaluated only on data from a specific region of interest, thus raising concerns about possible lack of generalization to other regions.<\/jats:p><\/jats:sec><jats:sec><jats:title>Purpose<\/jats:title><jats:p>In this work, we aim to address these limitations and propose LIRE: a learned invertible primal\u2010dual iterative scheme for CBCT\u00a0reconstruction.<\/jats:p><\/jats:sec><jats:sec><jats:title>Methods<\/jats:title><jats:p>LIRE is a learned invertible primal\u2010dual iterative scheme for CBCT reconstruction, wherein we employ a U\u2010Net architecture in each primal block and a residual convolutional neural network (CNN) architecture in each dual block. Memory requirements of the network are substantially reduced while preserving its expressive power through a combination of invertible residual primal\u2010dual blocks and patch\u2010wise computations inside each of the blocks during both forward and backward pass. These techniques enable us to train on data with isotropic 2 mm voxel spacing, clinically\u2010relevant projection count and detector panel resolution on current hardware with 24 GB\u00a0video random access memory (VRAM).<\/jats:p><\/jats:sec><jats:sec><jats:title>Results<\/jats:title><jats:p>Two LIRE models for small and for large field\u2010of\u2010view (FoV)\u00a0setting were trained and validated on a set of 260 + 22 thorax CT scans and tested using a set of 142 thorax CT scans plus an out\u2010of\u2010distribution dataset of 79 head and neck CT scans. For both settings, our method surpasses the classical methods and the deep learning baselines on both test sets. On the thorax CT set, our method achieves peak signal\u2010to\u2010noise ratio (PSNR) of 33.84 \u00b1 2.28 for the small FoV setting and 35.14 \u00b1 2.69 for the large FoV setting; U\u2010Net baseline achieves PSNR of 33.08 \u00b1 1.75 and 34.29 \u00b1 2.71 respectively. On the head and neck CT set, our method achieves PSNR of 39.35 \u00b1 1.75 for the small FoV setting and 41.21 \u00b1 1.41 for the large FoV setting; U\u2010Net baseline achieves PSNR of 33.08 \u00b1 1.75 and 34.29 \u00b1 2.71 respectively. Additionally, we demonstrate that LIRE can be finetuned to reconstruct high\u2010resolution CBCT data with the same geometry but 1 mm voxel spacing and higher detector panel resolution, where it outperforms the U\u2010Net baseline as\u00a0well.<\/jats:p><\/jats:sec><jats:sec><jats:title>Conclusions<\/jats:title><jats:p>Learned invertible primal\u2010dual schemes with additional memory optimizations can be trained to reconstruct CBCT volumes directly from the projection data with clinically\u2010relevant geometry and resolution. Such methods can offer better reconstruction quality and generalization compared to classical deep learning baselines.<\/jats:p><\/jats:sec>","DOI":"10.1002\/mp.16779","type":"journal-article","created":{"date-parts":[[2023,10,17]],"date-time":"2023-10-17T12:50:51Z","timestamp":1697547051000},"page":"7579-7593","update-policy":"https:\/\/doi.org\/10.1002\/crossmark_policy","source":"Crossref","is-referenced-by-count":11,"title":["End\u2010to\u2010end memory\u2010efficient reconstruction for cone beam\u00a0CT"],"prefix":"10.1002","volume":"50","author":[{"given":"Nikita","family":"Moriakov","sequence":"first","affiliation":[{"name":"Department of Radiation Oncology Netherlands Cancer Institute Amsterdam Netherlands"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jan\u2010Jakob","family":"Sonke","sequence":"additional","affiliation":[{"name":"Department of Radiation Oncology Netherlands Cancer Institute Amsterdam Netherlands"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jonas","family":"Teuwen","sequence":"additional","affiliation":[{"name":"Department of Radiation Oncology Netherlands Cancer Institute Amsterdam Netherlands"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"311","published-online":{"date-parts":[[2023,10,17]]},"reference":[{"key":"e_1_2_7_2_1","doi-asserted-by":"publisher","DOI":"10.1038\/sj.bdj.2009.560"},{"key":"e_1_2_7_3_1","doi-asserted-by":"publisher","DOI":"10.1007\/s11547-014-0429-5"},{"key":"e_1_2_7_4_1","doi-asserted-by":"publisher","DOI":"10.1016\/S0360-3016(02)02884-5"},{"key":"e_1_2_7_5_1","doi-asserted-by":"crossref","unstructured":"Maa\u00dfC DennerleinF NooF Kachelrie\u00dfM.Comparing short scan CT reconstruction algorithms regarding cone\u2010beam artifact performance. In:IEEE Nuclear Science Symposuim Medical Imaging Conference.IEEE;2010:2188\u20102193.","DOI":"10.1109\/NSSMIC.2010.5874170"},{"key":"e_1_2_7_6_1","doi-asserted-by":"publisher","DOI":"10.1137\/0143035"},{"key":"e_1_2_7_7_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.semradonc.2019.02.007"},{"key":"e_1_2_7_8_1","doi-asserted-by":"publisher","DOI":"10.1109\/TMI.2021.3075856"},{"key":"e_1_2_7_9_1","doi-asserted-by":"crossref","unstructured":"BeauferrisY TeuwenJ KarkalousosD et\u00a0al.Multi\u2010channel MR Reconstruction (MC\u2010MRRec) Challenge \u2013 Comparing Accelerated MR Reconstruction Models and Assessing Their Genereralizability to Datasets Collected with Different Coils.2020. doi:10.3389\/fnins.2022.919186","DOI":"10.3389\/fnins.2022.919186"},{"key":"e_1_2_7_10_1","doi-asserted-by":"publisher","DOI":"10.1017\/S0962492919000059"},{"key":"e_1_2_7_11_1","doi-asserted-by":"publisher","DOI":"10.1109\/TIP.2017.2713099"},{"key":"e_1_2_7_12_1","doi-asserted-by":"publisher","DOI":"10.1109\/TMI.2021.3081824"},{"key":"e_1_2_7_13_1","doi-asserted-by":"publisher","DOI":"10.1109\/TCI.2020.3019647"},{"key":"e_1_2_7_14_1","unstructured":"LunzS \u00d6ktemO Sch\u00f6nliebCB.Adversarial regularizers in inverse problems.Advances in Neural Information Processing Systems 31 (NeurIPS 2018).2018.arXiv e\u2010prints arXiv:1805.11572.https:\/\/papers.nips.cc\/paper_files\/paper\/2018\/hash\/d903e9608cfbf08910611e4346a0ba44\u2010Abstract.html"},{"key":"e_1_2_7_15_1","doi-asserted-by":"publisher","DOI":"10.1109\/TMI.2018.2829896"},{"key":"e_1_2_7_16_1","doi-asserted-by":"publisher","DOI":"10.1109\/TMI.2018.2799231"},{"key":"e_1_2_7_17_1","doi-asserted-by":"publisher","DOI":"10.1007\/s10851-010-0251-1"},{"key":"e_1_2_7_18_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.media.2021.102061"},{"key":"e_1_2_7_19_1","doi-asserted-by":"publisher","DOI":"10.3390\/app10051816"},{"key":"e_1_2_7_20_1","doi-asserted-by":"publisher","DOI":"10.1002\/mp.14170"},{"key":"e_1_2_7_21_1","doi-asserted-by":"publisher","DOI":"10.1109\/TCI.2020.2990299"},{"key":"e_1_2_7_22_1","author":"Rudzusika J","year":"2021","journal-title":"Invertible Learned Primal\u2010Dual"},{"key":"e_1_2_7_23_1","unstructured":"Baji\u0107B \u00d6ktemO RudzusikaJ.3D helical CT reconstruction with memory efficient invertible Learned Primal\u2010Dual method.2022.Arxiv preprint.https:\/\/arxiv.org\/abs\/2205.11952"},{"key":"e_1_2_7_24_1","doi-asserted-by":"publisher","DOI":"10.1007\/b138659"},{"key":"e_1_2_7_25_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.radonc.2005.03.001"},{"key":"e_1_2_7_26_1","doi-asserted-by":"publisher","DOI":"10.1002\/mp.13627"},{"key":"e_1_2_7_27_1","doi-asserted-by":"publisher","DOI":"10.1364\/JOSAA.1.000612"},{"key":"e_1_2_7_28_1","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-319-46723-8_49"},{"key":"e_1_2_7_29_1","doi-asserted-by":"crossref","unstructured":"WangZ CunX BaoJ ZhouW LiuJ LiH.Uformer: a general U\u2010shaped transformer for image restoration. In:Proceedings of the IEEE\/CVF Conference on Computer Vision and Pattern Recognition (CVPR).IEEE;2022:17683\u201017693.","DOI":"10.1109\/CVPR52688.2022.01716"},{"key":"e_1_2_7_30_1","unstructured":"AdlerJ KohrH \u00d6ktemO.ODL 0.6.0 2017. doi:10.5281\/zenodo.249479"},{"key":"e_1_2_7_31_1","doi-asserted-by":"publisher","DOI":"10.1118\/1.1450132"},{"key":"e_1_2_7_32_1","doi-asserted-by":"publisher","DOI":"10.1118\/1.4876724"},{"key":"e_1_2_7_33_1","doi-asserted-by":"publisher","DOI":"10.1118\/1.1489043"},{"key":"e_1_2_7_34_1","doi-asserted-by":"publisher","DOI":"10.3389\/fonc.2022.988800"},{"key":"e_1_2_7_35_1","doi-asserted-by":"publisher","DOI":"10.1002\/mp.16562"},{"key":"e_1_2_7_36_1","unstructured":"KingmaDP BaJ.Adam: a method for stochastic optimization. Published as a conference paper at the 3rd International Conference for Learning Representations 2015. doi:10.5281\/zenodo.249479"},{"key":"e_1_2_7_37_1","volume-title":"Advances in Neural Information Processing Systems","author":"Gomez AN","year":"2017"},{"key":"e_1_2_7_38_1","doi-asserted-by":"publisher","DOI":"10.1109\/TPAMI.2020.3019563"},{"key":"e_1_2_7_39_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.media.2019.01.005"},{"key":"e_1_2_7_40_1","first-page":"8024","volume-title":"Advances in Neural Information Processing Systems 32","author":"Paszke A","year":"2019"},{"key":"e_1_2_7_41_1","doi-asserted-by":"publisher","DOI":"10.1364\/OE.24.025129"},{"key":"e_1_2_7_42_1","doi-asserted-by":"publisher","DOI":"10.1002\/mp.13753"},{"key":"e_1_2_7_43_1","doi-asserted-by":"publisher","DOI":"10.1016\/0024-3795(74)90029-9"},{"key":"e_1_2_7_44_1","doi-asserted-by":"publisher","DOI":"10.21105\/joss.01576"},{"key":"e_1_2_7_45_1","series-title":"International Society for Optics and Photonics","first-page":"393","volume-title":"Medical Imaging 2018: Physics of Medical Imaging","author":"Maier J","year":"2018"}],"container-title":["Medical Physics"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/aapm.onlinelibrary.wiley.com\/doi\/pdf\/10.1002\/mp.16779","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2024,6,27]],"date-time":"2024-06-27T20:24:56Z","timestamp":1719519896000},"score":1,"resource":{"primary":{"URL":"https:\/\/aapm.onlinelibrary.wiley.com\/doi\/10.1002\/mp.16779"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2023,10,17]]},"references-count":44,"journal-issue":{"issue":"12","published-print":{"date-parts":[[2023,12]]}},"alternative-id":["10.1002\/mp.16779"],"URL":"https:\/\/doi.org\/10.1002\/mp.16779","archive":["Portico"],"relation":{},"ISSN":["0094-2405","2473-4209"],"issn-type":[{"value":"0094-2405","type":"print"},{"value":"2473-4209","type":"electronic"}],"subject":[],"published":{"date-parts":[[2023,10,17]]},"assertion":[{"value":"2023-02-07","order":0,"name":"received","label":"Received","group":{"name":"publication_history","label":"Publication History"}},{"value":"2023-08-08","order":1,"name":"accepted","label":"Accepted","group":{"name":"publication_history","label":"Publication History"}},{"value":"2023-10-17","order":2,"name":"published","label":"Published","group":{"name":"publication_history","label":"Publication History"}}]}}