{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,8,17]],"date-time":"2026-08-17T14:50:15Z","timestamp":1786978215950,"version":"build-2736575974"},"reference-count":37,"publisher":"Wiley","issue":"7","license":[{"start":{"date-parts":[[2024,2,20]],"date-time":"2024-02-20T00:00:00Z","timestamp":1708387200000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["aapm.onlinelibrary.wiley.com"],"crossmark-restriction":true},"short-container-title":["Medical Physics"],"published-print":{"date-parts":[[2024,7]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:sec>\n                    <jats:title>Background<\/jats:title>\n                    <jats:p>Auto\u2010segmentation of organs\u2010at\u2010risk (OARs) in the head and neck (HN) on computed tomography (CT) images is a time\u2010consuming component of the radiation therapy pipeline that suffers from inter\u2010observer variability. Deep learning (DL) has shown state\u2010of\u2010the\u2010art results in CT auto\u2010segmentation, with larger and more diverse datasets showing better segmentation performance. Institutional CT auto\u2010segmentation datasets have been small historically (n\u00a0&lt;\u00a050) due to the time required for manual curation of images and anatomical labels. Recently, large public CT auto\u2010segmentation datasets (n\u00a0&gt;\u00a01000 aggregated) have become available through online repositories such as The Cancer Imaging Archive. Transfer learning is a technique applied when training samples are scarce, but a large dataset from a closely related domain is available.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Purpose<\/jats:title>\n                    <jats:p>The purpose of this study was to investigate whether a large public dataset could be used in place of an institutional dataset (n\u00a0&gt;\u00a0500), or to augment performance via transfer learning, when building HN OAR auto\u2010segmentation models for institutional use.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>\n                      Auto\u2010segmentation models were trained on a large public dataset (public models) and a smaller institutional dataset (institutional models). The public models were fine\u2010tuned on the institutional dataset using transfer learning (transfer models). We assessed both public model generalizability and transfer model performance by comparison with institutional models. Additionally, the effect of institutional dataset size on both transfer and institutional models was investigated. All DL models used a high\u2010resolution, two\u2010stage architecture based on the popular 3D U\u2010Net. Model performance was evaluated using five geometric measures: the dice similarity coefficient (DSC), surface DSC, 95\n                      <jats:sup>th<\/jats:sup>\n                      percentile Hausdorff distance, mean surface distance (MSD), and added path length.\n                    <\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>\n                      For a small subset of OARs (left\/right optic nerve, spinal cord, left submandibular), the public models performed significantly better (\n                      <jats:italic>p<\/jats:italic>\n                      \u00a0&lt;\u00a00.05) than, or showed no significant difference to, the institutional models under most of the metrics examined. For the remaining OARs, the public models were inferior to the institutional models, although performance differences were small (DSC\u00a0\u2264\u00a00.03, MSD\u00a0&lt;\u00a00.5\u00a0mm) for seven OARs (brainstem, left\/right lens, left\/right parotid, mandible, right submandibular). The transfer models performed significantly better than the institutional models for seven OARs (brainstem, right lens, left\/right optic nerve, left\/right parotid, spinal cord) with a small margin of improvement (DSC\u00a0\u2264\u00a00.02, MSD\u00a0&lt;\u00a00.4\u00a0mm). When numbers of institutional training samples were limited, public and transfer models outperformed the institutional models for most OARs (brainstem, left\/right lens, left\/right optic nerve, left\/right parotid, spinal cord, and left\/right submandibular).\n                    <\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Conclusion<\/jats:title>\n                    <jats:p>Training auto\u2010segmentation models with public data alone was suitable for a small number of OARs. Using only public data incurred a small performance deficit for most other OARs, when compared with institutional data alone, but may be preferable over time\u2010consuming curation of a large institutional dataset. When a large institutional dataset was available, transfer learning with models pretrained on a large public dataset provided a modest performance improvement for several OARs. When numbers of institutional samples were limited, using the public dataset alone, or as a pretrained model, was beneficial for most OARs.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.1002\/mp.16997","type":"journal-article","created":{"date-parts":[[2024,2,20]],"date-time":"2024-02-20T10:38:01Z","timestamp":1708425481000},"page":"4767-4777","update-policy":"https:\/\/doi.org\/10.1002\/crossmark_policy","source":"Crossref","is-referenced-by-count":10,"title":["Transfer learning for auto\u2010segmentation of 17 organs\u2010at\u2010risk in the head and neck: Bridging the gap between institutional and public datasets"],"prefix":"10.1002","volume":"51","author":[{"given":"Brett","family":"Clark","sequence":"first","affiliation":[{"name":"Department of Biomedical Engineering University of Melbourne  Melbourne Australia"},{"name":"Department of Physical Sciences Peter MacCallum Cancer Centre  Melbourne Australia"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Nicholas","family":"Hardcastle","sequence":"additional","affiliation":[{"name":"Department of Physical Sciences Peter MacCallum Cancer Centre  Melbourne Australia"},{"name":"Centre for Medical Radiation Physics University of Wollongong  Wollongong Australia"},{"name":"Sir Peter MacCallum Department of Oncology University of Melbourne  Melbourne Australia"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Leigh A.","family":"Johnston","sequence":"additional","affiliation":[{"name":"Department of Biomedical Engineering University of Melbourne  Melbourne Australia"},{"name":"Melbourne Brain Centre Imaging Unit University of Melbourne  Melbourne Australia"},{"name":"Graeme Clark Institute University of Melbourne  Melbourne Australia"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"James","family":"Korte","sequence":"additional","affiliation":[{"name":"Department of Biomedical Engineering University of Melbourne  Melbourne Australia"},{"name":"Department of Physical Sciences Peter MacCallum Cancer Centre  Melbourne Australia"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"311","published-online":{"date-parts":[[2024,2,20]]},"reference":[{"key":"e_1_2_10_2_1","doi-asserted-by":"publisher","DOI":"10.1002\/cncr.21084"},{"key":"e_1_2_10_3_1","doi-asserted-by":"publisher","DOI":"10.1016\/S1470\u20102045(10)70290\u20104"},{"key":"e_1_2_10_4_1","doi-asserted-by":"publisher","DOI":"10.1200\/JCO.2015.63.9898"},{"key":"e_1_2_10_5_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.radonc.2015.07.041"},{"key":"e_1_2_10_6_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.radonc.2019.03.004"},{"key":"e_1_2_10_7_1","doi-asserted-by":"publisher","DOI":"10.1002\/mp.12045"},{"key":"e_1_2_10_8_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.phro.2020.06.006"},{"key":"e_1_2_10_9_1","doi-asserted-by":"publisher","DOI":"10.1117\/1.JMI.6.1.011005"},{"key":"e_1_2_10_10_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.radonc.2019.10.019"},{"key":"e_1_2_10_11_1","unstructured":"ChenS MaK ZhengY.Med3D: Transfer Learning for 3D Medical Image Analysis.ArXiv190400625 Cs. Published online July 17 2019. Accessed January 27 2021.http:\/\/arxiv.org\/abs\/1904.00625"},{"key":"e_1_2_10_12_1","unstructured":"SunC ShrivastavaA SinghS GuptaA.Revisiting Unreasonable Effectiveness of Data in Deep Learning Era.ArXiv170702968 Cs. Published online August 3 2017. Accessed March 12 2021.http:\/\/arxiv.org\/abs\/1707.02968"},{"key":"e_1_2_10_13_1","doi-asserted-by":"publisher","DOI":"10.1186\/s41747\u2010020\u201000173\u20102"},{"key":"e_1_2_10_14_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.radonc.2019.09.022"},{"key":"e_1_2_10_15_1","doi-asserted-by":"publisher","DOI":"10.1109\/JPROC.2020.3004555"},{"key":"e_1_2_10_16_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.artmed.2021.102078"},{"key":"e_1_2_10_17_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.ijrobp.2021.07.550"},{"key":"e_1_2_10_18_1","unstructured":"WeeL DekkerA.Data from Head\u2010Neck\u2010Radiomics\u2010HN1. Published online 2019. doi:10.7937\/TCIA.2019.8KAP372N"},{"key":"e_1_2_10_19_1","unstructured":"Valli\u00e8resM Kay\u2010RivestE PerrinL et\u00a0al.Data from Head\u2010Neck\u2010PET\u2010CT. Published online 2017. doi:10.7937\/K9\/TCIA.2017.8OJE5Q00"},{"key":"e_1_2_10_20_1","unstructured":"GrossbergA MohamedA El HalawaniH et\u00a0al.Data from Head and Neck Cancer CT Atlas. Published online 2017. doi:10.7937\/K9\/TCIA.2017.UMZ8DV6S"},{"key":"e_1_2_10_21_1","unstructured":"KwanJYY SuJ HuangSH et\u00a0al.Data from Radiomic Biomarkers to Refine Risk Models for Distant Metastasis in Oropharyngeal Carcinoma. Published online 2019. doi:10.7937\/TCIA.2019.8DHO2GLS"},{"key":"e_1_2_10_22_1","doi-asserted-by":"publisher","DOI":"10.1007\/s10278\u2010013\u20109622\u20107"},{"key":"e_1_2_10_23_1","doi-asserted-by":"publisher","DOI":"10.1109\/TKDE.2009.191"},{"key":"e_1_2_10_24_1","unstructured":"\u00c7i\u00e7ek\u00d6 AbdulkadirA LienkampSS BroxT.RonnebergerO 3D U\u2010Net: Learning Dense Volumetric Segmentation from Sparse Annotation.ArXiv160606650 Cs. Published online June 21 2016. Accessed November 4 2020.http:\/\/arxiv.org\/abs\/1606.06650"},{"key":"e_1_2_10_25_1","doi-asserted-by":"publisher","DOI":"10.1007\/s10278\u2010017\u20100037\u20108"},{"key":"e_1_2_10_26_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.cmpb.2021.106236"},{"key":"e_1_2_10_27_1","doi-asserted-by":"crossref","unstructured":"MilletariF NavabN AhmadiS.V\u2010Net: fully convolutional neural networks for volumetric medical image segmentation.2016 Fourth International Conference on 3D Vision (3DV).2016:565\u2010571. doi:10.1109\/3DV.2016.79","DOI":"10.1109\/3DV.2016.79"},{"key":"e_1_2_10_28_1","unstructured":"FalconWA.The PyTorch Lightning team. Pytorch lightning.GitHub. Published online 2019.https:\/\/github.com\/Lightning\u2010AI\/lightning"},{"key":"e_1_2_10_29_1","first-page":"32","article-title":"Pytorch: an imperative style, high\u2010performance deep learning library","author":"Paszke A","year":"2019","journal-title":"Adv Neural Inf Process Syst"},{"key":"e_1_2_10_30_1","unstructured":"NikolovS BlackwellS ZverovitchA et\u00a0al.Deep learning to achieve clinically applicable segmentation of head and neck anatomy for radiotherapy.ArXiv180904430 Phys Stat. Published online October 20 2020. Accessed November 3 2020.http:\/\/arxiv.org\/abs\/1809.04430"},{"key":"e_1_2_10_31_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.phro.2019.12.001"},{"key":"e_1_2_10_32_1","doi-asserted-by":"publisher","DOI":"10.1007\/s10278\u2010021\u201000460\u20103"},{"key":"e_1_2_10_33_1","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-319-66179-7_59"},{"key":"e_1_2_10_34_1","doi-asserted-by":"publisher","DOI":"10.1002\/mp.14320"},{"key":"e_1_2_10_35_1","unstructured":"OktayO SchlemperJ FolgocLL et\u00a0al.Attention U\u2010Net: Learning Where to Look for the Pancreas. Published online May 20 2018. doi:10.48550\/arXiv.1804.03999"},{"key":"e_1_2_10_36_1","unstructured":"ChenJ LuY YuQ et\u00a0al.TransUNet: Transformers Make Strong Encoders for Medical Image Segmentation. Published online February 8 2021. doi:10.48550\/arXiv.2102.04306"},{"key":"e_1_2_10_37_1","doi-asserted-by":"publisher","DOI":"10.1002\/mp.16197"},{"key":"e_1_2_10_38_1","doi-asserted-by":"publisher","DOI":"10.2307\/1932409"}],"container-title":["Medical Physics"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/aapm.onlinelibrary.wiley.com\/doi\/pdf\/10.1002\/mp.16997","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2024,7,9]],"date-time":"2024-07-09T11:56:38Z","timestamp":1720526198000},"score":1,"resource":{"primary":{"URL":"https:\/\/aapm.onlinelibrary.wiley.com\/doi\/10.1002\/mp.16997"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2024,2,20]]},"references-count":37,"journal-issue":{"issue":"7","published-print":{"date-parts":[[2024,7]]}},"alternative-id":["10.1002\/mp.16997"],"URL":"https:\/\/doi.org\/10.1002\/mp.16997","archive":["Portico"],"relation":{},"ISSN":["0094-2405","2473-4209"],"issn-type":[{"value":"0094-2405","type":"print"},{"value":"2473-4209","type":"electronic"}],"subject":[],"published":{"date-parts":[[2024,2,20]]},"assertion":[{"value":"2023-05-04","order":0,"name":"received","label":"Received","group":{"name":"publication_history","label":"Publication History"}},{"value":"2024-01-31","order":1,"name":"accepted","label":"Accepted","group":{"name":"publication_history","label":"Publication History"}},{"value":"2024-02-20","order":2,"name":"published","label":"Published","group":{"name":"publication_history","label":"Publication History"}}]}}