{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,8,24]],"date-time":"2026-08-24T05:19:02Z","timestamp":1787548742862,"version":"build-2736575974"},"reference-count":47,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,8,24]],"date-time":"2026-08-24T00:00:00Z","timestamp":1787529600000},"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. Oncol."],"abstract":"<jats:sec>\n                    <jats:title>Background<\/jats:title>\n                    <jats:p>Glioblastoma (GBM) exhibits profound molecular and spatial heterogeneity, complicating prognostic evaluations. While multiparametric MRI provides crucial multidimensional biological information, conventional end-to-end deep learning integration strategies, such as early or late fusion, often fail to capture complex nonlinear cross-modal interactions. We aimed to systematically evaluate a cross-attention fusion (CAF) architecture for GBM survival prediction and quantify its incremental prognostic value relative to existing clinical tools.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>In this multicenter retrospective study, 386 adults with IDH-wildtype, WHO grade 4 GBM were assembled from an institutional cohort (n = 226), the Chinese Glioma Genome Atlas (CGGA, n = 62), and The Cancer Genome Atlas (TCGA, n = 98). Using a unified 3D ResNet-18 backbone, we compared single-modality models, early fusion, late fusion, and CAF on preoperative T1-weighted, contrast-enhanced T1-weighted (T1CE), and T2-weighted MRI, and integrated the resulting deep learning risk score with routine clinical variables through multivariable Cox regression. Performance was assessed using Harrell\u2019s C-index, time-dependent AUC, and decision curve analysis.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>\n                      CAF showed numerically higher, more consistent C-index trends than early fusion, late fusion, and single-modality models (pooled C-index 0.629, 95% CI 0.594\u20130.664), although pairwise differences in time-dependent AUC were not statistically significant. Integrating clinical variables raised the pooled C-index to 0.691 (95% CI 0.660\u20130.721) in the treatment-era model, with comparable performance across the three cohorts (Local 0.688; CGGA 0.716; TCGA 0.689); a pre-treatment configuration excluding adjuvant therapy yielded a pooled C-index of 0.642. Under leave-one-cohort-out external validation, the combined model retained significant risk stratification in all held-out cohorts (C-index 0.63\u20130.71; all log-rank\n                      <jats:italic>P<\/jats:italic>\n                      \u00a0&amp;lt;\u00a00.01), albeit with attenuated discrimination. The deep learning risk score remained independent after multivariable adjustment (HR 1.41 per SD, 95% CI 1.26\u20131.57;\n                      <jats:italic>P<\/jats:italic>\n                      \u00a0&amp;lt;\u00a00.001). Kaplan\u2013Meier analysis confirmed significant high- versus low-risk separation in all cohorts, and decision curve analysis showed greater net benefit than clinical-only and deep-learning-only models.\n                    <\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Conclusion<\/jats:title>\n                    <jats:p>The CAF-derived risk score offers prognostic information complementary to routine clinical variables, representing a promising noninvasive tool for individualized risk stratification when molecular profiling is incomplete or unavailable; these findings warrant prospective external validation before clinical use.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.3389\/fonc.2026.1867582","type":"journal-article","created":{"date-parts":[[2026,8,24]],"date-time":"2026-08-24T04:23:38Z","timestamp":1787545418000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["Deep learning-based cross-attention fusion of multimodal MRI for survival prediction and risk stratification in IDH-wildtype glioblastoma: a multicenter study"],"prefix":"10.3389","volume":"16","author":[{"given":"Shichao","family":"Liu","sequence":"first","affiliation":[{"name":"Department of Neurosurgery, Fujian Medical University Union Hospital","place":["Fuzhou, Fujian, China"]},{"name":"Fujian Institute of Neurosurgery","place":["Fuzhou, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Risheng","family":"Liang","sequence":"additional","affiliation":[{"name":"Department of Neurosurgery, Fujian Medical University Union Hospital","place":["Fuzhou, Fujian, China"]},{"name":"Fujian Institute of Neurosurgery","place":["Fuzhou, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,8,24]]},"reference":[{"key":"B1","doi-asserted-by":"publisher","DOI":"10.1093\/neuonc\/noab106","article-title":"The 2021 WHO classification of tumors of the central nervous system: a summary","volume":"23","author":"Louis","year":"2021","journal-title":"Neuro-Oncol"},{"key":"B2","doi-asserted-by":"publisher","DOI":"10.1056\/nejmoa043330","article-title":"Radiotherapy plus concomitant and adjuvant temozolomide for glioblastoma","volume":"352","author":"Stupp","year":"2005","journal-title":"N Engl J Med"},{"key":"B3","doi-asserted-by":"publisher","DOI":"10.1097\/WCO.0000000000001320","article-title":"Updates for newly diagnosed and recurrent glioblastoma: a review of recent clinical trials","volume":"37","author":"Fukushima","year":"2024","journal-title":"Curr Opin Neurol"},{"key":"B4","doi-asserted-by":"publisher","DOI":"10.3171\/2024.5.JNS24393","article-title":"Long-term survivors in 976 supratentorial glioblastoma, IDH-wildtype patients","volume":"142","author":"Aboubakr","year":"2025","journal-title":"J Neurosurg"},{"key":"B5","doi-asserted-by":"publisher","first-page":"iv1","DOI":"10.1093\/neuonc\/noad149","article-title":"CBTRUS statistical report: primary brain and other central nervous system tumors diagnosed in the United States in 2016-2020","volume":"25","author":"Ostrom","year":"2023","journal-title":"Neuro-Oncol"},{"key":"B6","doi-asserted-by":"publisher","DOI":"10.1093\/neuonc\/now208","article-title":"An independently validated nomogram for individualized estimation of survival among patients with newly diagnosed glioblastoma: NRG oncology RTOG 0525 and 0825","volume":"19","author":"Gittleman","year":"2017","journal-title":"Neuro-Oncol"},{"key":"B7","doi-asserted-by":"publisher","first-page":"36","DOI":"10.1186\/s43046-025-00296-w","article-title":"IDH mutation and MGMT methylation status in glioblastoma and other gliomas patients: a Russian retrospective cohort study","volume":"37","author":"Eid","year":"2025","journal-title":"J Egypt Natl Cancer Inst"},{"key":"B8","doi-asserted-by":"publisher","first-page":"997","DOI":"10.1056\/NEJMoa043331","article-title":"MGMT gene silencing and benefit from temozolomide in glioblastoma","volume":"352","author":"Hegi","year":"2005","journal-title":"N Engl J Med"},{"key":"B9","doi-asserted-by":"publisher","DOI":"10.1016\/j.compbiomed.2024.109436","article-title":"Deep learning-based overall survival prediction in patients with glioblastoma: an automatic end-to-end workflow using pre-resection basic structural multiparametric MRIs","volume":"185","author":"Yang","year":"2025","journal-title":"Comput Biol Med"},{"key":"B10","doi-asserted-by":"publisher","DOI":"10.1109\/isbi60581.2025.10981128","article-title":"Overall survival prediction of brain tumor patients with multimodal mri using swin unetr","volume":"2025","author":"Kim","year":"2025","journal-title":"Proc IEEE Int Symp BioMed Imaging"},{"key":"B11","doi-asserted-by":"publisher","DOI":"10.1093\/noajnl\/vdae122","article-title":"Comprehensive multimodal deep learning survival prediction enabled by a transformer architecture: a multicenter study in glioblastoma","volume":"6","author":"Gomaa","year":"2024","journal-title":"Neuro-Oncol Adv"},{"key":"B12","doi-asserted-by":"publisher","first-page":"1505","DOI":"10.1186\/s12885-025-14906-2","article-title":"Survival risk stratification of 2021 WHO glioblastoma by MRI radiomics and biological exploration","volume":"25","author":"Li","year":"2025","journal-title":"BMC Cancer"},{"key":"B13","doi-asserted-by":"publisher","first-page":"8784","DOI":"10.1038\/s41598-022-12699-z","article-title":"Clinical measures, radiomics, and genomics offer synergistic value in AI-based prediction of overall survival in patients with glioblastoma","volume":"12","author":"Fathi Kazerooni","year":"2022","journal-title":"Sci Rep"},{"key":"B14","doi-asserted-by":"publisher","first-page":"136","DOI":"10.1038\/s41746-020-00341-z","article-title":"Fusion of medical imaging and electronic health records using deep learning: a systematic review and implementation guidelines","volume":"3","author":"Huang","year":"2020","journal-title":"NPJ Dig Med"},{"key":"B15","doi-asserted-by":"publisher","DOI":"10.1126\/science.1254257","article-title":"Single-cell RNA-seq highlights intratumoral heterogeneity in primary glioblastoma","volume":"344","author":"Patel","year":"2014","journal-title":"Science"},{"key":"B16","doi-asserted-by":"publisher","first-page":"98","DOI":"10.1016\/j.ccr.2009.12.020","article-title":"Integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA, IDH1, EGFR, and NF1","volume":"17","author":"Verhaak","year":"2010","journal-title":"Cancer Cell"},{"key":"B17","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1016\/j.gpb.2020.10.005","article-title":"Chinese glioma genome atlas (CGGA): a comprehensive resource with functional genomic data from chinese glioma patients","volume":"19","author":"Zhao","year":"2021","journal-title":"Genomics Proteomics Bioinf"},{"key":"B18","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2107.02314","article-title":"The RSNA-ASNR-MICCAI BraTS 2021 benchmark on brain tumor segmentation and radiogenomic classification","author":"Baid","year":"2021"},{"key":"B19","doi-asserted-by":"publisher","DOI":"10.1109\/TMI.2010.2046908","article-title":"N4ITK: improved N3 bias correction","volume":"29","author":"Tustison","year":"2010","journal-title":"IEEE Trans Med Imag"},{"key":"B20","doi-asserted-by":"publisher","first-page":"26","DOI":"10.1016\/j.media.2007.06.004","article-title":"Symmetric diffeomorphic image registration with cross-correlation: evaluating automated labeling of elderly and neurodegenerative brain","volume":"12","author":"Avants","year":"2008","journal-title":"Med Img Anal"},{"key":"B21","doi-asserted-by":"publisher","DOI":"10.1038\/s41592-020-01008-z","article-title":"nnU-net: a self-configuring method for deep learning-based biomedical image segmentation","volume":"18","author":"Isensee","year":"2021","journal-title":"Nat Methods"},{"key":"B22","doi-asserted-by":"publisher","first-page":"170117","DOI":"10.1038\/sdata.2017.117","article-title":"Advancing the cancer genome atlas glioma MRI collections with expert segmentation labels and radiomic features","volume":"4","author":"Bakas","year":"2017","journal-title":"Sci Data"},{"key":"B23","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1904.00625","article-title":"Med3D: transfer learning for 3D medical image analysis","author":"Chen","year":"2019"},{"key":"B24","doi-asserted-by":"publisher","DOI":"10.1503\/cmaj.110977","article-title":"Missing covariate data in clinical research: when and when not to use the missing-indicator method for analysis","volume":"184","author":"Groenwold","year":"2012","journal-title":"CMAJ: Can Med Assoc J = J Assoc Med Can"},{"key":"B25","doi-asserted-by":"publisher","DOI":"10.1002\/(SICI)1097-0258(19960229)15:43C361::AID-SIM1683E3.0.CO;2-4","article-title":"Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors","volume":"15","author":"Harrell","year":"1996","journal-title":"Stat Med"},{"key":"B26","doi-asserted-by":"publisher","DOI":"10.1177\/0272989X06295361","article-title":"Decision curve analysis: a novel method for evaluating prediction models","volume":"26","author":"Vickers","year":"2006","journal-title":"Med Decis Mak: Int J Soc Med Decis Mak"},{"key":"B27","doi-asserted-by":"publisher","DOI":"10.2307\/2531595","article-title":"Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach","volume":"44","author":"DeLong","year":"1988","journal-title":"Biometrics"},{"key":"B28","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1802.03426","article-title":"UMAP: uniform manifold approximation and projection for dimension reduction","author":"McInnes","year":"2020"},{"key":"B29","doi-asserted-by":"publisher","DOI":"10.1007\/s11263-019-01228-7","article-title":"Grad-CAM: visual explanations from deep networks via gradient-based localization","volume":"128","author":"Selvaraju","year":"2020","journal-title":"Int J Comput Vision"},{"key":"B30","doi-asserted-by":"publisher","first-page":"797","DOI":"10.1148\/radiol.2018180200","article-title":"Radiomic MRI phenotyping of glioblastoma: improving survival prediction","volume":"289","author":"Bae","year":"2018","journal-title":"Radiology"},{"key":"B31","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-319-46723-8_25","article-title":"3D deep learning for multi-modal imaging-guided survival time prediction of brain tumor patients","volume":"9901","author":"Nie","year":"2016","journal-title":"Med Img Comput Comput-assist Interv: MICCAI Int Conf Med Img Comput Comput-Assist Interv"},{"key":"B32","doi-asserted-by":"publisher","DOI":"10.3390\/cancers12082284","article-title":"Multi-parametric deep learning model for prediction of overall survival after postoperative concurrent chemoradiotherapy in glioblastoma patients","volume":"12","author":"Yoon","year":"2020","journal-title":"Cancers"},{"key":"B33","doi-asserted-by":"publisher","DOI":"10.1007\/s00330-022-08640-7","article-title":"Glioma survival prediction from whole-brain MRI without tumor segmentation using deep attention network: a multicenter study","volume":"32","author":"Li","year":"2022","journal-title":"Eur Radio"},{"key":"B34","doi-asserted-by":"publisher","DOI":"10.1007\/s11060-023-04439-8","article-title":"Predicting survival in glioblastoma with multimodal neuroimaging and machine learning","volume":"164","author":"Luckett","year":"2023","journal-title":"J Neuro-Oncol"},{"key":"B35","doi-asserted-by":"publisher","DOI":"10.1093\/neuonc\/noae260","article-title":"Machine learning-based prognostic subgrouping of glioblastoma: a multicenter study","volume":"27","author":"Akbari","year":"2025","journal-title":"Neuro-Oncol"},{"key":"B36","doi-asserted-by":"publisher","DOI":"10.1007\/s00109-013-1019-z","article-title":"Mechanisms of neovascularization and resistance to anti-angiogenic therapies in glioblastoma multiforme","volume":"91","author":"Soda","year":"2013","journal-title":"J Mol Med"},{"key":"B37","doi-asserted-by":"publisher","DOI":"10.1117\/1.JMI.5.2.021219","article-title":"Radiomic signature of infiltration in peritumoral edema predicts subsequent recurrence in glioblastoma: implications for personalized radiotherapy planning","volume":"5","author":"Rathore","year":"2018","journal-title":"J Med Imaging"},{"key":"B38","doi-asserted-by":"publisher","DOI":"10.1007\/s00330-016-4637-3","article-title":"Radiomic features from the peritumoral brain parenchyma on treatment-na\u00efve multi-parametric MR imaging predict long versus short-term survival in glioblastoma multiforme: preliminary findings","volume":"27","author":"Prasanna","year":"2017","journal-title":"Eur Radio"},{"key":"B39","doi-asserted-by":"publisher","DOI":"10.18632\/oncotarget.22947","article-title":"Spatial habitats from multiparametric MR imaging are associated with signaling pathway activities and survival in glioblastoma","volume":"8","author":"Dextraze","year":"2017","journal-title":"Oncotarget"},{"key":"B40","doi-asserted-by":"publisher","DOI":"10.3390\/cancers12071707","article-title":"Multi-habitat radiomics unravels distinct phenotypic subtypes of glioblastoma with clinical and genomic significance","volume":"12","author":"Choi","year":"2020","journal-title":"Cancers"},{"key":"B41","doi-asserted-by":"publisher","DOI":"10.1016\/j.acra.2024.09.021","article-title":"Deep learning and habitat radiomics for the prediction of glioma pathology using multiparametric MRI: a multicenter study","volume":"32","author":"Zhu","year":"2025","journal-title":"Acad Radiol"},{"key":"B42","doi-asserted-by":"publisher","first-page":"55","DOI":"10.1038\/s43856-025-00767-0","article-title":"The radiogenomic and spatiogenomic landscapes of glioblastoma and their relationship to oncogenic drivers","volume":"5","author":"Fathi Kazerooni","year":"2025","journal-title":"Commun Med"},{"key":"B43","doi-asserted-by":"publisher","DOI":"10.1093\/neuonc\/nox175","article-title":"Is the blood-brain barrier really disrupted in all glioblastomas? A critical assessment of existing clinical data","volume":"20","author":"Sarkaria","year":"2018","journal-title":"Neuro-Oncol"},{"key":"B44","doi-asserted-by":"publisher","DOI":"10.1093\/jnci\/85.9.704","article-title":"Recursive partitioning analysis of prognostic factors in three radiation therapy oncology group Malignant glioma trials","volume":"85","author":"Curran","year":"1993","journal-title":"J Natl Cancer Inst"},{"key":"B45","doi-asserted-by":"publisher","first-page":"63","DOI":"10.1038\/s41698-025-01269-x","article-title":"AI-powered segmentation and prognosis with missing MRI in pediatric brain tumors","volume":"10","author":"Chrysochoou","year":"2026","journal-title":"NPJ Precis Oncol"},{"key":"B46","doi-asserted-by":"publisher","DOI":"10.1007\/s00330-024-10728-1","article-title":"Medical image foundation models in assisting diagnosis of brain tumors: a pilot study","volume":"34","author":"Chen","year":"2024","journal-title":"Eur Radio"},{"key":"B47","doi-asserted-by":"publisher","DOI":"10.1038\/s41593-026-02202-6","article-title":"A generalizable foundation model for analysis of human brain MRI","volume":"29","author":"Tak","year":"2026","journal-title":"Nat Neurosci"}],"container-title":["Frontiers in Oncology"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fonc.2026.1867582\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,8,24]],"date-time":"2026-08-24T04:23:40Z","timestamp":1787545420000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fonc.2026.1867582\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,8,24]]},"references-count":47,"alternative-id":["10.3389\/fonc.2026.1867582"],"URL":"https:\/\/doi.org\/10.3389\/fonc.2026.1867582","relation":{},"ISSN":["2234-943X"],"issn-type":[{"value":"2234-943X","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,8,24]]},"article-number":"1867582"}}