{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,16]],"date-time":"2026-10-16T12:44:07Z","timestamp":1792154647913,"version":"4.4.0"},"reference-count":36,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,9,15]],"date-time":"2026-09-15T00:00:00Z","timestamp":1789430400000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Artif. Intell."],"abstract":"<jats:sec>\n                    <jats:title>Background<\/jats:title>\n                    <jats:p>Accurate preoperative assessment of pulmonary nodule invasiveness remains challenging. We developed an internally validated multimodal framework integrating CT representations from a frozen vision foundation model with clinical variables.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>\n                      This retrospective single-centre study included 1,179 pathologically confirmed pulmonary nodules: 247 glandular precursor lesions comprising atypical adenomatous hyperplasia and adenocarcinoma\n                      <jats:italic>in situ<\/jats:italic>\n                      , and 932 invasive lesions comprising minimally invasive and invasive adenocarcinoma. CT volumes were resampled to 0.5-mm isotropic resolution and cropped into 64\u202f\u00d7\u202f64\u202f\u00d7\u202f64-voxel patches. Slice-level representations were extracted using a pretrained, frozen DINOv3 backbone and aggregated by a trainable Attention Probe. Encoded clinical variables and imaging representations were integrated through self-attention and bidirectional cross-attention, followed by neural classification and regression tree classification. Internal validation used a five-fold rotating train\u2013validation\u2013test procedure, with each fold serving once as the held-out test fold.\n                    <\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>\n                      The held-out test-fold AUCs were 0.848, 0.864, 0.867, 0.882, and 0.822, yielding a mean AUC of 0.8566. Pooled out-of-fold predictions produced an AUC of 0.847, accuracy of 0.809, sensitivity of 0.806, specificity of 0.822, and F1 score of 0.873. DINOv3 and NCART achieved the highest point-estimate AUCs among the evaluated feature extractors and classifiers, respectively, although most pairwise differences were not statistically significant. Intermediate fusion significantly outperformed the Gould score and the clinical-data-only model, but not the imaging-only or late-fusion models. In the prespecified secondary analysis, the model achieved an AUC of 0.780 for distinguishing adenocarcinoma\n                      <jats:italic>in situ<\/jats:italic>\n                      from atypical adenomatous hyperplasia.\n                    <\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Conclusion<\/jats:title>\n                    <jats:p>The proposed framework achieved internally validated discrimination of pulmonary nodule invasiveness. External multicentre and prospective validation is required before clinical implementation.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.3389\/frai.2026.1889248","type":"journal-article","created":{"date-parts":[[2026,9,15]],"date-time":"2026-09-15T05:36:08Z","timestamp":1789450568000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["Imaging-based development and validation of artificial intelligence models for lung adenocarcinoma precursor lesions and early lung adenocarcinoma presenting as pulmonary nodules"],"prefix":"10.3389","volume":"9","author":[{"given":"Junbao","family":"Zhang","sequence":"first","affiliation":[{"name":"Department of Pulmonary and Critical Care Medicine, Peking University Shenzhen Hospital","place":["Shenzhen, China"]},{"name":"Department of Pulmonary and Critical Care Medicine, Huashan Hospital, Fudan University","place":["Shanghai, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yanyi","family":"Hou","sequence":"additional","affiliation":[{"name":"Department of Pulmonary and Critical Care Medicine, Peking University Shenzhen Hospital","place":["Shenzhen, China"]},{"name":"Shenzhen University Medical School","place":["Shenzhen, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yikang","family":"Yang","sequence":"additional","affiliation":[{"name":"Shenzhen University Medical School","place":["Shenzhen, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ziyan","family":"Zhang","sequence":"additional","affiliation":[{"name":"Department of Radiotherapy Oncology, Sun Yat-sen Memorial Hospital, Sun Yat-sen University","place":["Guangzhou, China"]},{"name":"Department of Radiation Oncology, Sichuan Cancer Hospital and Institute, University of Electronic Science and Technology of China","place":["Chengdu, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yi","family":"Gao","sequence":"additional","affiliation":[{"name":"Shenzhen University Medical School","place":["Shenzhen, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ping","family":"Xu","sequence":"additional","affiliation":[{"name":"Department of Pulmonary and Critical Care Medicine, Peking University Shenzhen Hospital","place":["Shenzhen, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,9,15]]},"reference":[{"key":"ref1","doi-asserted-by":"publisher","first-page":"103381","DOI":"10.1016\/j.media.2024.103381","article-title":"A cross-attention-based deep learning approach for predicting functional stroke outcomes using 4D CTP imaging and clinical metadata","volume":"99","author":"Amador","year":"2025","journal-title":"Med. Image Anal."},{"key":"ref2","doi-asserted-by":"publisher","first-page":"229","DOI":"10.3322\/caac.21834","article-title":"Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries","volume":"74","author":"Bray","year":"2024","journal-title":"CA Cancer J. Clin."},{"key":"ref3","doi-asserted-by":"publisher","first-page":"97","DOI":"10.1016\/j.semcancer.2023.05.004","article-title":"Radiomics and artificial intelligence for precision medicine in lung cancer treatment","volume":"93","author":"Chen","year":"2023","journal-title":"Semin. Cancer Biol."},{"key":"ref4","first-page":"774","volume-title":"Diagnostic classification of lung nodules using 3D neural networks 2018 IEEE 15th international symposium on biomedical imaging (ISBI 2018)","author":"Dey","year":"2018"},{"key":"ref5","doi-asserted-by":"publisher","first-page":"BSR20212416","DOI":"10.1042\/BSR20212416","article-title":"Identification of pathological subtypes of early lung adenocarcinoma based on artificial intelligence parameters and CT signs","volume":"42","author":"Fang","year":"2022","journal-title":"Biosci. Rep."},{"key":"ref6","doi-asserted-by":"publisher","first-page":"24991","DOI":"10.52202\/068431-1812","article-title":"On embeddings for numerical features in tabular deep learning","volume":"35","author":"Gorishniy","year":"2022","journal-title":"Adv. Neural Inf. Process. Syst"},{"key":"ref7","doi-asserted-by":"publisher","first-page":"383","DOI":"10.1378\/chest.06-1261","article-title":"A clinical model to estimate the pretest probability of lung cancer in patients with solitary pulmonary nodules","volume":"131","author":"Gould","year":"2007","journal-title":"Chest"},{"key":"ref8","doi-asserted-by":"publisher","first-page":"736","DOI":"10.1038\/s41746-025-02041-y","article-title":"ShapeField-lung: continuous shape embedding for early lung cancer detection via pulmonary nodule segmentation","volume":"8","author":"Gu","year":"2025","journal-title":"NPJ Digit. Med."},{"key":"ref9","first-page":"15979","volume-title":"2022 IEEE\/CVF Conference on Computer Vision and Pattern Recognition (CVPR)","author":"He","year":"2021"},{"key":"ref10","first-page":"770","volume-title":"2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)","author":"He","year":"2015"},{"key":"ref11","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2012.06678","article-title":"TabTransformer: tabular data modeling using contextual embeddings","author":"Huang","year":"2020","journal-title":"ArXiv"},{"key":"ref12","doi-asserted-by":"publisher","first-page":"103","DOI":"10.4143\/crt.2022.264","article-title":"Five-year overall survival and prognostic factors in patients with lung Cancer: results from the Korean Association of Lung Cancer Registry (KALC-R) 2015","volume":"55","author":"Jeon","year":"2023","journal-title":"Cancer Res. Treat."},{"key":"ref13","doi-asserted-by":"publisher","first-page":"116137","DOI":"10.1016\/j.ejca.2025.116137","article-title":"AI performance for nodule volume doubling time in the follow-up of the UKLS lung cancer screening study compared to expert consensus and histological validation","volume":"232","author":"Jiang","year":"2026","journal-title":"Eur. J. Cancer"},{"key":"ref14","doi-asserted-by":"publisher","first-page":"873","DOI":"10.1148\/radiol.2015141071","article-title":"Solitary pure ground-glass nodules 5 mm or smaller: frequency of growth","volume":"276","author":"Kakinuma","year":"2015","journal-title":"Radiology"},{"key":"ref15","doi-asserted-by":"publisher","first-page":"311","DOI":"10.3348\/jksr.2022.0093","article-title":"Hematoma-filled Pneumatocele after CT-guided percutaneous transthoracic needle lung biopsy: two case reports","volume":"84","author":"Kang","year":"2023","journal-title":"J Korean Soc Radiol"},{"key":"ref16","doi-asserted-by":"publisher","first-page":"2245","DOI":"10.1177\/02841851221144616","article-title":"Effect of the respiratory motion of pulmonary nodules on CT-guided percutaneous transthoracic needle biopsy","volume":"64","author":"Kim","year":"2023","journal-title":"Acta Radiol."},{"key":"ref17","doi-asserted-by":"publisher","first-page":"749","DOI":"10.1038\/nrclinonc.2017.141","article-title":"Radiomics: the bridge between medical imaging and personalized medicine","volume":"14","author":"Lambin","year":"2017","journal-title":"Nat. Rev. Clin. Oncol."},{"key":"ref18","doi-asserted-by":"publisher","first-page":"441","DOI":"10.1016\/j.ejca.2011.11.036","article-title":"Radiomics: extracting more information from medical images using advanced feature analysis","volume":"48","author":"Lambin","year":"2012","journal-title":"Eur. J. Cancer"},{"key":"ref19","doi-asserted-by":"publisher","first-page":"624","DOI":"10.1038\/s41571-023-00798-3","article-title":"The global burden of lung cancer: current status and future trends","volume":"20","author":"Leiter","year":"2023","journal-title":"Nat. Rev. Clin. Oncol."},{"key":"ref20","doi-asserted-by":"publisher","first-page":"108","DOI":"10.1111\/ajco.14042","article-title":"Overdiagnosis and overtreatment of ground-glass nodule-like lung cancer","volume":"21","author":"Liang","year":"2025","journal-title":"Asia Pac. J. Clin. Oncol."},{"key":"ref21","doi-asserted-by":"publisher","first-page":"e836","DOI":"10.1002\/ctm2.836","article-title":"A novel EHD1\/CD44\/hippo\/SP1 positive feedback loop potentiates stemness and metastasis in lung adenocarcinoma","volume":"12","author":"Liu","year":"2022","journal-title":"Clin. Transl. Med."},{"key":"ref22","doi-asserted-by":"publisher","first-page":"228","DOI":"10.1148\/radiol.2017161659","article-title":"Guidelines for Management of Incidental Pulmonary Nodules Detected on CT images: from the Fleischner society 2017","volume":"284","author":"MacMahon","year":"2017","journal-title":"Radiology"},{"key":"ref23","doi-asserted-by":"publisher","first-page":"409","DOI":"10.3322\/caac.21731","article-title":"Cancer treatment and survivorship statistics, 2022","volume":"72","author":"Miller","year":"2022","journal-title":"CA Cancer J. Clin."},{"key":"ref24","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2304.07193","article-title":"DINOv2: learning robust visual features without supervision","author":"Oquab","year":"2023","journal-title":"ArXiv"},{"key":"ref25","doi-asserted-by":"publisher","first-page":"517","DOI":"10.1016\/j.rcl.2024.12.002","article-title":"Imaging and Management of Subsolid Lung Nodules","volume":"63","author":"Raad","year":"2025","journal-title":"Radiol. Clin. North Am."},{"key":"ref26","doi-asserted-by":"publisher","first-page":"2335","DOI":"10.21037\/tlcr-20-296","article-title":"Screen-detected solid nodules: from detection of nodule to structured reporting","volume":"10","author":"Silva","year":"2021","journal-title":"Transl Lung Cancer Res"},{"key":"ref27","author":"Sim\u00e9oni","year":"2025"},{"key":"ref28","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1409.1556","article-title":"Very deep convolutional networks for large-scale image recognition","author":"Simonyan","year":"2014","journal-title":"arXiv"},{"key":"ref29","doi-asserted-by":"publisher","first-page":"e1","DOI":"10.1016\/j.jtho.2021.09.017","article-title":"Introduction to 2021 WHO classification of thoracic tumors","volume":"17","author":"Tsao","year":"2022","journal-title":"J. Thorac. Oncol."},{"key":"ref30","first-page":"5998","volume-title":"Attention Is All You Need. Advances in Neural Information Processing Systems","author":"Vaswani","year":"2017"},{"key":"ref31","doi-asserted-by":"publisher","first-page":"130","DOI":"10.1701\/3315.32853","article-title":"Radiomics and artificial intelligence: new frontiers in medicine","volume":"111","author":"Vernuccio","year":"2020","journal-title":"Recenti Prog. Med."},{"key":"ref32","doi-asserted-by":"publisher","first-page":"3184","DOI":"10.1038\/s41591-024-03211-3","article-title":"Data-driven risk stratification and precision management of pulmonary nodules detected on chest computed tomography","volume":"30","author":"Wang","year":"2024","journal-title":"Nat. Med."},{"key":"ref33","doi-asserted-by":"publisher","first-page":"1937","DOI":"10.1097\/CM9.0000000000002729","article-title":"Persistent increase and improved survival of stage I lung cancer based on a large-scale real-world sample of 26,226 cases","volume":"136","author":"Wang","year":"2023","journal-title":"Chin. Med. J."},{"key":"ref34","doi-asserted-by":"publisher","first-page":"2416799","DOI":"10.1016\/j.pulmoe.2024.02.010","article-title":"Personalised follow-up and management schema for patients with screen-detected pulmonary nodules: a dynamic modelling study","volume":"31","author":"Wang","year":"2025","journal-title":"Pulmonology"},{"key":"ref35","volume-title":"Thoracic Tumours. WHO Classification of Tumours","year":"2021"},{"key":"ref36","doi-asserted-by":"publisher","first-page":"76","DOI":"10.1136\/thorax-2024-221642","article-title":"Diameter thresholds for pure ground-glass pulmonary nodules at low-dose CT screening: Chinese experience","volume":"80","author":"Ye","year":"2025","journal-title":"Thorax"}],"container-title":["Frontiers in Artificial Intelligence"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1889248\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,10,16]],"date-time":"2026-10-16T11:46:23Z","timestamp":1792151183000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1889248\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,15]]},"references-count":36,"alternative-id":["10.3389\/frai.2026.1889248"],"URL":"https:\/\/doi.org\/10.3389\/frai.2026.1889248","relation":{},"ISSN":["2624-8212"],"issn-type":[{"value":"2624-8212","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,9,15]]},"article-number":"1889248"}}