{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,29]],"date-time":"2026-09-29T10:06:47Z","timestamp":1790676407521,"version":"4.1.0"},"reference-count":31,"publisher":"Wiley","issue":"10","license":[{"start":{"date-parts":[[2023,4,29]],"date-time":"2023-04-29T00:00:00Z","timestamp":1682726400000},"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":[[2023,10]]},"abstract":"<jats:title>Abstract<\/jats:title><jats:sec><jats:title>Background<\/jats:title><jats:p>Clinical data used to train deep learning models are often not clean data. They can contain imperfections in both the imaging data and the corresponding segmentations.<\/jats:p><\/jats:sec><jats:sec><jats:title>Purpose<\/jats:title><jats:p>This study investigates the influence of data imperfections on the performance of deep learning models for parotid gland segmentation. This was done in a controlled manner by using synthesized data. The insights this study provides may be used to make deep learning models better and more reliable.<\/jats:p><\/jats:sec><jats:sec><jats:title>Methods<\/jats:title><jats:p>The data were synthesized by using the clinical segmentations, creating a pseudo ground\u2010truth in the process. Three kinds of imperfections were simulated: incorrect segmentations, low image contrast, and artifacts in the imaging data. The severity of each imperfection was varied in five levels. Models resulting from training sets from each of the five levels were cross\u2010evaluated with test sets from each of the five levels.<\/jats:p><\/jats:sec><jats:sec><jats:title>Results<\/jats:title><jats:p>Using synthesized data led to almost perfect parotid gland segmentation when no error was added. Lowering the quality of the parotid gland segmentations used for training substantially lowered the model performance. Additionally, lowering the image quality of the training data by decreasing the contrast or introducing artifacts made the resulting models more robust to data containing those respective kinds of data imperfection.<\/jats:p><\/jats:sec><jats:sec><jats:title>Conclusion<\/jats:title><jats:p>This study demonstrated the importance of good\u2010quality segmentations for deep learning training and it shows that using low\u2010quality imaging data for training can enhance the robustness of the resulting models.<\/jats:p><\/jats:sec>","DOI":"10.1002\/mp.16437","type":"journal-article","created":{"date-parts":[[2023,4,29]],"date-time":"2023-04-29T07:23:58Z","timestamp":1682753038000},"page":"6421-6432","update-policy":"https:\/\/doi.org\/10.1002\/crossmark_policy","source":"Crossref","is-referenced-by-count":16,"title":["Impact of imperfection in medical imaging data on deep learning\u2010based segmentation performance: An experimental study using synthesized data"],"prefix":"10.1002","volume":"50","author":[{"given":"Ayetullah Mehdi","family":"G\u00fcne\u015f","sequence":"first","affiliation":[{"name":"Department of Radiation Oncology Amsterdam UMC  Amsterdam The Netherlands"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ward","family":"van Rooij","sequence":"additional","affiliation":[{"name":"Department of Radiation Oncology Amsterdam UMC  Amsterdam The Netherlands"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Sadaf","family":"Gulshad","sequence":"additional","affiliation":[{"name":"Faculty of Science Universiteit van Amsterdam  Amsterdam The Netherlands"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ben","family":"Slotman","sequence":"additional","affiliation":[{"name":"Department of Radiation Oncology Amsterdam UMC  Amsterdam The Netherlands"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Max","family":"Dahele","sequence":"additional","affiliation":[{"name":"Department of Radiation Oncology Amsterdam UMC  Amsterdam The Netherlands"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Wilko","family":"Verbakel","sequence":"additional","affiliation":[{"name":"Department of Radiation Oncology Amsterdam UMC  Amsterdam The Netherlands"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"311","published-online":{"date-parts":[[2023,4,29]]},"reference":[{"key":"e_1_2_8_2_1","doi-asserted-by":"publisher","DOI":"10.1186\/1748717X\u20107\u201032"},{"key":"e_1_2_8_3_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.radonc.2015.07.041"},{"key":"e_1_2_8_4_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.ijrobp.2010.10.019"},{"key":"e_1_2_8_5_1","doi-asserted-by":"publisher","DOI":"10.1007\/s10278-019-00227-x"},{"key":"e_1_2_8_6_1","doi-asserted-by":"publisher","DOI":"10.1117\/1.JMI.6.1.011005"},{"key":"e_1_2_8_7_1","unstructured":"NikolovS BlackwellS MendesR et\u00a0al. Deep learning to achieve clinically applicable segmentation of head and neck anatomy for radiotherapy;2018. arXiv:1809.04430"},{"key":"e_1_2_8_8_1","doi-asserted-by":"publisher","DOI":"10.1002\/mp.13300"},{"key":"e_1_2_8_9_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.radonc.2019.05.010"},{"key":"e_1_2_8_10_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.ijrobp.2019.02.040"},{"key":"e_1_2_8_11_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.phro.2020.06.006"},{"key":"e_1_2_8_12_1","doi-asserted-by":"publisher","DOI":"10.1053\/srao.2002.31359"},{"key":"e_1_2_8_13_1","doi-asserted-by":"publisher","DOI":"10.1002\/cncr.22302"},{"key":"e_1_2_8_14_1","doi-asserted-by":"publisher","DOI":"10.1118\/1.4747526"},{"key":"e_1_2_8_15_1","unstructured":"LongJ ShelhamerE DarrellT. Fully convolutional networks for semantic segmentation;2014. arXiv:1411.4038"},{"key":"e_1_2_8_16_1","doi-asserted-by":"crossref","unstructured":"CicekO AbdulkadirA LienkampSS et\u00a0al. 3D U\u2010Net: learning dense volumetric segmentation from sparse annotation;2016. arXiv:1606.06650v1","DOI":"10.1007\/978-3-319-46723-8_49"},{"key":"e_1_2_8_17_1","first-page":"1929","article-title":"Dropout: a simple way to prevent neural networks from overfitting","volume":"15","author":"Srivastava N","year":"2014","journal-title":"J Mach Learn Res."},{"key":"e_1_2_8_18_1","unstructured":"KingmaDP BaJ. Adam: a method for stochastic optimization;2015. arXiv:1412.6980v9"},{"key":"e_1_2_8_19_1","unstructured":"ReddiSJ KaleS andKumarS. On the convergence of Adam and beyond;2019. arXiv:1904.09237"},{"key":"e_1_2_8_20_1","doi-asserted-by":"publisher","DOI":"10.1118\/s13014\u2010020\u201001721\u20101"},{"key":"e_1_2_8_21_1","doi-asserted-by":"crossref","unstructured":"SimmardPY SteinkrausD andPlattJC. Best practices for convolutional neural networks applied to visual document analysis:Proceedings of the Seventh International Conference on Document Analysis and Recognition;2003. doi:10.1109\/ICDAR.2003.1227801","DOI":"10.1109\/ICDAR.2003.1227801"},{"key":"e_1_2_8_22_1","doi-asserted-by":"publisher","DOI":"10.1088\/1361\u20106560\/ab99e5"},{"key":"e_1_2_8_23_1","unstructured":"RolnickD VeitA BelongieS et\u00a0al. Deep learning is robust to massive label noise;2017. arXiv:1705.10694."},{"key":"e_1_2_8_24_1","unstructured":"ArpitD Jastrz\u0119bskiS BallasN et\u00a0al. A closer look at memorization in deep networks;2017. arXiv:1706.05394"},{"key":"e_1_2_8_25_1","unstructured":"VasiljevicI ChakrabartiA andShakhnarovichG. Examining the impact of blur on recognition by convolutional networks;2017. arXiv:1611.05760"},{"key":"e_1_2_8_26_1","doi-asserted-by":"publisher","DOI":"10.3389\/fncom.2019.00083"},{"key":"e_1_2_8_27_1","doi-asserted-by":"publisher","DOI":"10.1613\/jair.295"},{"key":"e_1_2_8_28_1","unstructured":"WardD MoghadamP andHudsonN. Deep leaf segmentation using synthetic data;2018. arXiv:1807.10931"},{"key":"e_1_2_8_29_1","doi-asserted-by":"crossref","unstructured":"PhanB TruongS KhanR et\u00a0al. Bayesian uncertainty quantification with synthetic data. WAISE 2019: second international workshop on artificial intelligence safety engineering;2019.","DOI":"10.1007\/978-3-030-26250-1_31"},{"key":"e_1_2_8_30_1","doi-asserted-by":"crossref","unstructured":"SankaranarayananS BalajiY JainA et\u00a0al. Learning from synthetic data: addressing domain shift for semantic segmentation;2018. arXiv:1711.06969","DOI":"10.1109\/CVPR.2018.00395"},{"key":"e_1_2_8_31_1","doi-asserted-by":"crossref","unstructured":"ChenY LiW ChenX et\u00a0al. Learning semantic segmentation from synthetic data: a geometrically guided input\u2010output adaptation approach;2019. arXiv:1812.05040","DOI":"10.1109\/CVPR.2019.00194"},{"key":"e_1_2_8_32_1","doi-asserted-by":"publisher","DOI":"10.1002\/mp.13697"}],"container-title":["Medical Physics"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/aapm.onlinelibrary.wiley.com\/doi\/pdf\/10.1002\/mp.16437","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2023,10,12]],"date-time":"2023-10-12T05:45:11Z","timestamp":1697089511000},"score":1,"resource":{"primary":{"URL":"https:\/\/aapm.onlinelibrary.wiley.com\/doi\/10.1002\/mp.16437"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2023,4,29]]},"references-count":31,"journal-issue":{"issue":"10","published-print":{"date-parts":[[2023,10]]}},"alternative-id":["10.1002\/mp.16437"],"URL":"https:\/\/doi.org\/10.1002\/mp.16437","archive":["Portico"],"relation":{},"ISSN":["0094-2405","2473-4209"],"issn-type":[{"value":"0094-2405","type":"print"},{"value":"2473-4209","type":"electronic"}],"subject":[],"published":{"date-parts":[[2023,4,29]]},"assertion":[{"value":"2022-03-12","order":0,"name":"received","label":"Received","group":{"name":"publication_history","label":"Publication History"}},{"value":"2023-03-12","order":1,"name":"accepted","label":"Accepted","group":{"name":"publication_history","label":"Publication History"}},{"value":"2023-04-29","order":2,"name":"published","label":"Published","group":{"name":"publication_history","label":"Publication History"}}]}}