{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,20]],"date-time":"2026-10-20T16:41:35Z","timestamp":1792514495410,"version":"4.5.0"},"reference-count":52,"publisher":"Wiley","issue":"11","license":[{"start":{"date-parts":[[2026,10,20]],"date-time":"2026-10-20T00:00:00Z","timestamp":1792454400000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by\/4.0\/"},{"start":{"date-parts":[[2026,10,20]],"date-time":"2026-10-20T00:00:00Z","timestamp":1792454400000},"content-version":"tdm","delay-in-days":0,"URL":"http:\/\/doi.wiley.com\/10.1002\/tdm_license_1.1"}],"content-domain":{"domain":["aapm.onlinelibrary.wiley.com"],"crossmark-restriction":true},"short-container-title":["J Applied Clin Med Phys"],"published-print":{"date-parts":[[2026,11]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:sec>\n                    <jats:title>Background<\/jats:title>\n                    <jats:p>Report\u2010mined chest radiograph datasets commonly contain both long\u2010tailed label distributions and uncertain annotations. Conventional frequency\u2010based re\u2010weighting can strengthen rare\u2010label learning, but it may also amplify unreliable supervision when uncertain labels occur in minority classes.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Purpose<\/jats:title>\n                    <jats:p>To develop and evaluate an uncertainty\u2010suppressed long\u2010tailed learning method with lightweight feature adaptation for multi\u2010label chest X\u2010ray (CXR) classification under imbalanced and uncertain report\u2010derived supervision.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>The proposed uncertainty\u2010suppressed long\u2010tailed feature adaptation (USLFA) method combines a lightweight label\u2010adaptive feature head with an uncertainty\u2010suppressed long\u2010tailed loss. The primary analysis used a patient\u2010level 70%\/10%\/20% split of CheXpert for 14 observation labels. Robustness was further examined across two additional patient\u2010level partitions, on the recommended MIMIC\u2010CXR\u2010JPG test split without retraining, and on the official radiologist\u2010annotated CheXpert test set. Performance was assessed using macro\u2010AUROC, macro\u2010AUPRC, macro F1, frequency\u2010group analysis, ablation studies, and uncertainty\u2010handling comparisons. Ninety\u2010five percent confidence intervals (CIs) were estimated by patient\u2010level bootstrap resampling, and formal pairwise tests in the primary CheXpert analysis used paired permutation testing with Holm correction.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>\n                      On the primary CheXpert split, USLFA achieved a macro\u2010AUROC of 0.864 (95% CI, 0.856\u20130.871), macro\u2010AUPRC of 0.547 (95% CI, 0.532\u20130.560), and macro F1 of 0.501 (95% CI, 0.487\u20130.514). Compared with ConvNeXt\u2010Tiny using binary cross\u2010entropy (BCE), the gains in macro\u2010AUPRC and macro F1 were both 0.051 and remained significant after Holm correction (both\n                      <jats:italic>p\u00a0&lt;<\/jats:italic>\n                      \u00a00.001). Compared with label\u2010distribution\u2010aware margin (LDAM), USLFA improved macro\u2010AUPRC by 0.018 (Holm\u2010adjusted\n                      <jats:italic>p\u00a0=<\/jats:italic>\n                      \u00a00.024) and macro F1 by 0.017 (Holm\u2010adjusted\n                      <jats:italic>p\u00a0=<\/jats:italic>\n                      \u00a00.041). The ordering was preserved across three patient\u2010level partitions, with USLFA macro\u2010AUPRC ranging from 0.540 to 0.555. On MIMIC\u2010CXR\u2010JPG, USLFA achieved a macro\u2010AUPRC of 0.485 (95% CI, 0.467\u20130.502), and on the official CheXpert test set it achieved a mean AUROC of 0.920 (95% CI, 0.908\u20130.931) across the five competition observations.\n                    <\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Conclusions<\/jats:title>\n                    <jats:p>USLFA improved macro\u2010AUPRC and rare\u2010label recognition without introducing an additional inference\u2010time branch. Macro F1 was significantly higher than the principal BCE and LDAM baselines in the main comparison, whereas the additional macro F1 gain attributable to uncertainty suppression over U Soft was not significant after Holm correction. The repeated\u2010split, external\u2010dataset, and official\u2010test analyses provide complementary evidence that the observed advantage is not confined to a single custom partition.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.1002\/acm2.70873","type":"journal-article","created":{"date-parts":[[2026,10,20]],"date-time":"2026-10-20T15:56:12Z","timestamp":1792511772000},"update-policy":"https:\/\/doi.org\/10.1002\/crossmark_policy","source":"Crossref","is-referenced-by-count":0,"title":["Uncertainty\u2010suppressed long\u2010tailed learning with lightweight feature adaptation for multi\u2010label chest X\u2010ray classification"],"prefix":"10.1002","volume":"27","author":[{"ORCID":"https:\/\/orcid.org\/0009-0003-6611-7711","authenticated-orcid":false,"given":"Lu","family":"Li","sequence":"first","affiliation":[{"name":"Department of Radiology Hubei Province Academy of Traditional Chinese Medicine, Hubei Provincial Hospital of Traditional Chinese Medicine (Guanggu Campus)  Wuhan Hubei Province China"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"311","published-online":{"date-parts":[[2026,10,20]]},"reference":[{"key":"e_1_2_12_2_1","doi-asserted-by":"publisher","DOI":"10.1038\/s44401\u2010026\u201000087\u2010y"},{"key":"e_1_2_12_3_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.media.2025.103739"},{"key":"e_1_2_12_4_1","doi-asserted-by":"crossref","unstructured":"DongH LinY ZhouP et\u00a0al. \u201cOverview of the CXR\u2010LT 2026 Challenge: multi\u2010Center Long\u2010Tailed and Zero Shot Chest X\u2010Ray Classification \u201d inProceedings of the IEEE International Symposium on Biomedical Imaging2026;2026. doi:10.1109\/ISBI61048.2026.11515307","DOI":"10.1109\/ISBI61048.2026.11515307"},{"key":"e_1_2_12_5_1","doi-asserted-by":"crossref","unstructured":"IrvinJ RajpurkarP KoM et\u00a0al. \u201cCheXpert: a large chest radiograph dataset with uncertainty labels and expert comparison \u201d inProceedings of the AAAI Conference on Artificial Intelligence. 33:590\u2010597 2019.","DOI":"10.1609\/aaai.v33i01.3301590"},{"key":"e_1_2_12_6_1","unstructured":"JohnsonAEW LungrenMP PengY et\u00a0al. \u201cMIMIC\u2010CXR\u2010JPG\u2014chest radiographs with structured labels \u201dPhysioNet Version 2.1.0 2024. doi:10.13026\/jsn5\u2010t979"},{"key":"e_1_2_12_7_1","doi-asserted-by":"crossref","unstructured":"WangX PengY LuL LuZ BagheriM SummersRM \u201cChestX\u2010ray8: hospital\u2010scale chest X\u2010ray database and benchmarks on weakly\u2010supervised classification and localization of common thorax diseases \u201d inProceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition. 2097\u20102106 2017.","DOI":"10.1109\/CVPR.2017.369"},{"key":"e_1_2_12_8_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.media.2020.101797"},{"key":"e_1_2_12_9_1","doi-asserted-by":"publisher","DOI":"10.1038\/s41746\u2010025\u201002032\u2010z"},{"key":"e_1_2_12_10_1","doi-asserted-by":"publisher","DOI":"10.1109\/TMI.2025.3581907"},{"key":"e_1_2_12_11_1","volume-title":"A multimodal biomedical foundation model trained from fifteen million scientific image\u2010text pairs","author":"Zhang S","year":"2025"},{"key":"e_1_2_12_12_1","doi-asserted-by":"publisher","DOI":"10.1038\/s41598\u2010026\u201053115\u20100"},{"key":"e_1_2_12_13_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.eswa.2025.129898"},{"key":"e_1_2_12_14_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.bspc.2024.107018"},{"key":"e_1_2_12_15_1","doi-asserted-by":"publisher","DOI":"10.1038\/s41598\u2010025\u201016669\u2010z"},{"key":"e_1_2_12_16_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.engappai.2026.113873"},{"key":"e_1_2_12_17_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.bspc.2026.109930"},{"key":"e_1_2_12_18_1","doi-asserted-by":"publisher","DOI":"10.1109\/JBHI.2020.2967084"},{"key":"e_1_2_12_19_1","doi-asserted-by":"publisher","DOI":"10.1007\/s11517\u2010022\u201002604\u20101"},{"key":"e_1_2_12_20_1","doi-asserted-by":"publisher","DOI":"10.1109\/TMI.2024.3421644"},{"key":"e_1_2_12_21_1","doi-asserted-by":"crossref","unstructured":"ChoJ KimD KimM JangH LeeCS YeJC \u201cCXR\u2010LT 2026 Challenge: projection\u2010aware multi\u2010label and zero\u2010shot chest X\u2010ray classification \u201d inProceedings of the IEEE International Symposium on Biomedical Imaging 2026. doi:10.1109\/ISBI61048.2026.11515904","DOI":"10.1109\/ISBI61048.2026.11515904"},{"key":"e_1_2_12_22_1","doi-asserted-by":"crossref","unstructured":"CuiY JiaM LinT\u2010Y SongY BelongieS \u201cClass\u2010balanced loss based on effective number of samples \u201d inProceedings of the IEEE\/CVF Conference on Computer Vision and Pattern Recognition. 9268\u20109277 2019.","DOI":"10.1109\/CVPR.2019.00949"},{"key":"e_1_2_12_23_1","unstructured":"CaoK WeiC GaidonA ArechigaN MaT \u201cLearning imbalanced datasets with label\u2010distribution\u2010aware margin loss \u201d inAdvances in Neural Information Processing Systems 32 2019."},{"key":"e_1_2_12_24_1","doi-asserted-by":"crossref","unstructured":"RidnikT Ben\u2010BaruchE ZamirN et\u00a0al. \u201cAsymmetric loss for multi\u2010label classification \u201d inProceedings of the IEEE\/CVF International Conference on Computer Vision 82\u201091 2021.","DOI":"10.1109\/ICCV48922.2021.00015"},{"key":"e_1_2_12_25_1","unstructured":"LinT\u2010Y GoyalP GirshickR HeK DollarP \u201cFocal loss for dense object detection \u201d inProceedings of the IEEE International Conference on Computer Vision 2980\u20102988 2017."},{"key":"e_1_2_12_26_1","doi-asserted-by":"crossref","unstructured":"WuT HuangQ LiuZ WangY LinD \u201cDistribution\u2010balanced loss for multi\u2010label classification in long\u2010tailed datasets \u201d inProceedings of the European Conference on Computer Vision 162\u2010178 2020.","DOI":"10.1007\/978-3-030-58548-8_10"},{"key":"e_1_2_12_27_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.eswa.2026.131438"},{"key":"e_1_2_12_28_1","doi-asserted-by":"publisher","DOI":"10.1098\/rsta.2025.0129"},{"key":"e_1_2_12_29_1","doi-asserted-by":"crossref","unstructured":"SmitA JainS RajpurkarP PareekA NgAY LungrenMP \u201cCheXbert: combining automatic labelers and expert annotations for accurate radiology report labeling using BERT \u201d inProceedings of the Conference on Empirical Methods in Natural Language Processing 1500\u20101519 2020.","DOI":"10.18653\/v1\/2020.emnlp-main.117"},{"key":"e_1_2_12_30_1","unstructured":"ChambonP et\u00a0al. \u201cCheXpert Plus: augmenting a large chest X\u2010ray dataset with text radiology reports patient demographics and additional image formats \u201darXiv:2405.19538 2024."},{"key":"e_1_2_12_31_1","doi-asserted-by":"crossref","unstructured":"deCastroDC BustosA BannurS et\u00a0al. \u201cPadChest\u2010GR: a bilingual chest X\u2010ray dataset for grounded radiology report generation \u201dNEJM AI 2025. doi:10.1056\/AIdbp2401120","DOI":"10.1056\/AIdbp2401120"},{"key":"e_1_2_12_32_1","doi-asserted-by":"publisher","DOI":"10.1007\/s10278\u2010025\u201001494\u20107"},{"key":"e_1_2_12_33_1","doi-asserted-by":"publisher","DOI":"10.1038\/s41598\u2010025\u201095390\u20103"},{"key":"e_1_2_12_34_1","doi-asserted-by":"publisher","DOI":"10.3390\/diagnostics15222899"},{"key":"e_1_2_12_35_1","doi-asserted-by":"publisher","DOI":"10.1371\/journal.pmed.1002683"},{"key":"e_1_2_12_36_1","unstructured":"RajpurkarP IrvinJ ZhuK et\u00a0al. \u201cCheXNet: radiologist\u2010level pneumonia detection on chest X\u2010rays with deep learning \u201darXiv:1711.05225 2017."},{"key":"e_1_2_12_37_1","doi-asserted-by":"publisher","DOI":"10.1371\/journal.pmed.1002686"},{"key":"e_1_2_12_38_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.patrec.2018.10.027"},{"key":"e_1_2_12_39_1","doi-asserted-by":"crossref","unstructured":"HuangS\u2010C ShenL LungrenMP YeungS \u201cGLoRIA: a multimodal global\u2010local representation learning framework for label\u2010efficient medical image recognition \u201d inProceedings of the IEEE\/CVF International Conference on Computer Vision 3942\u20103951 2021.","DOI":"10.1109\/ICCV48922.2021.00391"},{"key":"e_1_2_12_40_1","doi-asserted-by":"publisher","DOI":"10.1038\/s41551-022-00936-9"},{"key":"e_1_2_12_41_1","doi-asserted-by":"crossref","unstructured":"WangZ WuZ AgarwalD SunJ \u201cMedCLIP: contrastive learning from unpaired medical images and text \u201d inProceedings of the Conference on Empirical Methods in Natural Language Processing 3876\u20103887 2022.","DOI":"10.18653\/v1\/2022.emnlp-main.256"},{"key":"e_1_2_12_42_1","unstructured":"StatherosJM WangH KleinR \u201cCLARiTy: a vision transformer for multi\u2010label classification and weakly\u2010supervised localization of chest X\u2010ray pathologies \u201darXiv:2512.16700 2025."},{"key":"e_1_2_12_43_1","unstructured":"RandA IbrahimH \u201cBeyond Conventional Transformers: the medical X\u2010ray attention (MXA) block for improved multi\u2010label diagnosis using knowledge distillation\u201d arXiv:2504.02277 2025."},{"key":"e_1_2_12_44_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.aej.2026.01.022"},{"key":"e_1_2_12_45_1","doi-asserted-by":"crossref","unstructured":"SelvarajuRR CogswellM DasA et\u00a0al. \u201cGrad\u2010CAM: visual explanations from deep networks via gradient\u2010based localization \u201d inProceedings of the IEEE International Conference on Computer Vision 618\u2010626 2017.","DOI":"10.1109\/ICCV.2017.74"},{"key":"e_1_2_12_46_1","doi-asserted-by":"publisher","DOI":"10.1148\/ryai.230060"},{"key":"e_1_2_12_47_1","doi-asserted-by":"publisher","DOI":"10.1371\/journal.pone.0118432"},{"key":"e_1_2_12_48_1","doi-asserted-by":"crossref","unstructured":"HuangG LiuZ van derMaatenL WeinbergerKQ \u201cDensely connected convolutional networks \u201d inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition 4700\u20104708 2017. doi:10.1109\/CVPR.2017.243","DOI":"10.1109\/CVPR.2017.243"},{"key":"e_1_2_12_49_1","doi-asserted-by":"crossref","unstructured":"LiuZ MaoH WuC\u2010Y FeichtenhoferC DarrellT XieS \u201cA ConvNet for the 2020s \u201d inProceedings of the IEEE\/CVF Conference on Computer Vision and Pattern Recognition 11966\u201011976 2022. doi:10.1109\/CVPR52688.2022.01167","DOI":"10.1109\/CVPR52688.2022.01167"},{"key":"e_1_2_12_50_1","doi-asserted-by":"crossref","unstructured":"DengJ DongW SocherR LiL\u2010J LiK Fei\u2010FeiL \u201cImageNet: a large\u2010scale hierarchical image database \u201d inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition 248\u2010255 2009. doi:10.1109\/CVPR.2009.5206848","DOI":"10.1109\/CVPR.2009.5206848"},{"key":"e_1_2_12_51_1","unstructured":"PaszkeA GrossS MassaF et\u00a0al. \u201cPyTorch: an imperative style high\u2010performance deep learning library \u201d inAdvances in Neural Information Processing Systems32:8024\u20108035 2019."},{"key":"e_1_2_12_52_1","unstructured":"LoshchilovI HutterF \u201cDecoupled weight decay regularization \u201d inProceedings of the International Conference on Learning Representations 2019."},{"key":"e_1_2_12_53_1","unstructured":"LoshchilovI HutterF \u201cSGDR: stochastic gradient descent with warm restarts \u201d inProceedings of the International Conference on Learning Representations 2017."}],"container-title":["Journal of Applied Clinical Medical Physics"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/aapm.onlinelibrary.wiley.com\/doi\/pdf\/10.1002\/acm2.70873","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/aapm.onlinelibrary.wiley.com\/doi\/full-xml\/10.1002\/acm2.70873","content-type":"application\/xml","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/aapm.onlinelibrary.wiley.com\/doi\/pdf\/10.1002\/acm2.70873","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,10,20]],"date-time":"2026-10-20T15:56:15Z","timestamp":1792511775000},"score":1,"resource":{"primary":{"URL":"https:\/\/aapm.onlinelibrary.wiley.com\/doi\/10.1002\/acm2.70873"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,10,20]]},"references-count":52,"journal-issue":{"issue":"11","published-print":{"date-parts":[[2026,11]]}},"alternative-id":["10.1002\/acm2.70873"],"URL":"https:\/\/doi.org\/10.1002\/acm2.70873","archive":["Portico"],"relation":{},"ISSN":["1526-9914","1526-9914"],"issn-type":[{"value":"1526-9914","type":"print"},{"value":"1526-9914","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,10,20]]},"assertion":[{"value":"2026-08-02","order":0,"name":"received","label":"Received","group":{"name":"publication_history","label":"Publication History"}},{"value":"2026-10-12","order":2,"name":"accepted","label":"Accepted","group":{"name":"publication_history","label":"Publication History"}},{"value":"2026-10-20","order":3,"name":"published","label":"Published","group":{"name":"publication_history","label":"Publication History"}}],"article-number":"e70873"}}