{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,8,7]],"date-time":"2026-08-07T17:07:12Z","timestamp":1786122432147,"version":"build-2736575974"},"reference-count":35,"publisher":"MDPI AG","issue":"16","license":[{"start":{"date-parts":[[2026,8,7]],"date-time":"2026-08-07T00:00:00Z","timestamp":1786060800000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"name":"National Institutes of Health, National Cancer Institute","award":["R21CA273665"],"award-info":[{"award-number":["R21CA273665"]}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["Cancers"],"abstract":"<jats:p>Background\/Objectives: Digital pathology models increasingly infer molecular phenotypes from hematoxylin and eosin whole-slide images (WSIs), but most assume that one slide adequately represents patient-level tumor biology. We quantified prediction instability caused by slide selection and evaluated multi-WSI aggregation. Methods: Six multiple-instance learning architectures were developed in task-specific subsets of TCGA-BRCA (source cohort: 1006 patients and 1065 WSIs) and independently evaluated in 122 patients (400 WSIs) from CPTAC-BRCA. The clinically realistic comparison was random one-slide-per-patient inference versus patient-level aggregation of all available WSIs. Label-conditioned best- and worst-slide analyses were used only as retrospective oracle bounds. The recurrence-risk endpoint was a research-derived 21-gene recurrence-score surrogate calculated from RNA sequencing rather than a clinically reported Oncotype DX result. Results: Random single-slide selection yielded AUCs of 0.77\u20130.86 for recurrence-risk prediction and 0.65\u20130.77 for HER2 status. Patient-level multi-WSI aggregation yielded AUCs of 0.81\u20130.89 and 0.71\u20130.80, respectively. The architecture-controlled AUC improvement from random selection to aggregation ranged from 0.026 to 0.041 for recurrence risk and from 0.036 to 0.073 for HER2. Aggregation also reduced Brier scores by 0.010\u20130.028 and 0.009\u20130.106, respectively. The label-conditioned oracle analyses demonstrated wider theoretical performance ranges of 0.46\u20130.97 and 0.40\u20130.92, respectively. Slide-selection sensitivity persisted after excluding tumor-evidence-negative and low-tumor-content WSIs. Conclusions: Slide selection is a material source of sampling-dependent prediction uncertainty. Patient-level multi-WSI aggregation mitigated selection-dependent degradation across architectures, although validation using routine institutional material and clinically reported molecular assays remains necessary.<\/jats:p>","DOI":"10.3390\/cancers18162540","type":"journal-article","created":{"date-parts":[[2026,8,7]],"date-time":"2026-08-07T10:12:01Z","timestamp":1786097521000},"page":"2540","update-policy":"https:\/\/doi.org\/10.3390\/mdpi_crossmark_policy","source":"Crossref","is-referenced-by-count":0,"title":["Slide Selection Introduces Sampling-Induced Prediction Uncertainty in Digital Pathology AI: Evidence from Multi-Slide Breast Cancer Cohorts"],"prefix":"10.3390","volume":"18","author":[{"ORCID":"https:\/\/orcid.org\/0000-0002-4755-3935","authenticated-orcid":false,"given":"Onur C.","family":"Koyun","sequence":"first","affiliation":[{"name":"Center for Artificial Intelligence Research, Wake Forest University School of Medicine, Winston-Salem, NC 27101, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yongxin","family":"Guo","sequence":"additional","affiliation":[{"name":"Center for Artificial Intelligence Research, Wake Forest University School of Medicine, Winston-Salem, NC 27101, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-4547-7780","authenticated-orcid":false,"given":"Hao","family":"Lu","sequence":"additional","affiliation":[{"name":"Center for Artificial Intelligence Research, Wake Forest University School of Medicine, Winston-Salem, NC 27101, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0004-3666-7525","authenticated-orcid":false,"given":"Muhammet Fatih","family":"Demir","sequence":"additional","affiliation":[{"name":"Center for Artificial Intelligence Research, Wake Forest University School of Medicine, Winston-Salem, NC 27101, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-7849-6881","authenticated-orcid":false,"given":"Abbas","family":"Alili","sequence":"additional","affiliation":[{"name":"Center for Artificial Intelligence Research, Wake Forest University School of Medicine, Winston-Salem, NC 27101, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0008-2111-5380","authenticated-orcid":false,"given":"Nabil","family":"Rahoui","sequence":"additional","affiliation":[{"name":"Department of Pathology, Wake Forest University School of Medicine, Winston-Salem, NC 27157, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Diana","family":"Cardona","sequence":"additional","affiliation":[{"name":"Department of Pathology, Wake Forest University School of Medicine, Winston-Salem, NC 27157, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-2421-8229","authenticated-orcid":false,"given":"Metin N.","family":"Gurcan","sequence":"additional","affiliation":[{"name":"Center for Artificial Intelligence Research, Wake Forest University School of Medicine, Winston-Salem, NC 27101, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1968","published-online":{"date-parts":[[2026,8,7]]},"reference":[{"key":"ref_1","unstructured":"National Comprehensive Cancer Network (2020). NCCN Guidelines for Patients: Invasive Breast Cancer, NCCN Foundation."},{"key":"ref_2","first-page":"171","article-title":"Breast cancer treatment","volume":"104","author":"Trayes","year":"2021","journal-title":"Am. Fam. Physician"},{"key":"ref_3","doi-asserted-by":"crossref","first-page":"e253","DOI":"10.1016\/S1470-2045(19)30154-8","article-title":"Digital pathology and artificial intelligence","volume":"20","author":"Niazi","year":"2019","journal-title":"Lancet Oncol."},{"key":"ref_4","doi-asserted-by":"crossref","first-page":"703","DOI":"10.1038\/s41571-019-0252-y","article-title":"Artificial intelligence in digital pathology\u2014New tools for diagnosis and precision oncology","volume":"16","author":"Bera","year":"2019","journal-title":"Nat. Rev. Clin. Oncol."},{"key":"ref_5","doi-asserted-by":"crossref","first-page":"1559","DOI":"10.1038\/s41591-018-0177-5","article-title":"Classification and mutation prediction from non-small cell lung cancer histopathology images using deep learning","volume":"24","author":"Coudray","year":"2018","journal-title":"Nat. Med."},{"key":"ref_6","doi-asserted-by":"crossref","first-page":"329","DOI":"10.1016\/j.patcog.2017.10.009","article-title":"Multiple instance learning: A survey of problem characteristics and applications","volume":"77","author":"Carbonneau","year":"2018","journal-title":"Pattern Recognit."},{"key":"ref_7","doi-asserted-by":"crossref","unstructured":"Omoush, S.A., Alzyoud, J.A., El-Omari, N.K.T., and Alzyoud, A.J. (2026). The role of whole slide imaging in AI-based digital pathology: Current challenges and future directions\u2014An updated literature review. J. Mol. Pathol., 7.","DOI":"10.3390\/jmp7010002"},{"key":"ref_8","unstructured":"Coleman, W.B., and Tsongalis, G.J. (2023). Diagnostic Molecular Pathology: A Guide to Applied Molecular Testing, Academic Press."},{"key":"ref_9","doi-asserted-by":"crossref","first-page":"210","DOI":"10.1186\/bcr3658","article-title":"Breast cancer intra-tumor heterogeneity","volume":"16","author":"Martelotto","year":"2014","journal-title":"Breast Cancer Res."},{"key":"ref_10","doi-asserted-by":"crossref","first-page":"883","DOI":"10.1056\/NEJMoa1113205","article-title":"Intratumor heterogeneity and branched evolution revealed by multiregion sequencing","volume":"366","author":"Gerlinger","year":"2012","journal-title":"N. Engl. J. Med."},{"key":"ref_11","doi-asserted-by":"crossref","first-page":"323","DOI":"10.1038\/nrc3261","article-title":"Intra-tumour heterogeneity: A looking glass for cancer?","volume":"12","author":"Marusyk","year":"2012","journal-title":"Nat. Rev. Cancer"},{"key":"ref_12","doi-asserted-by":"crossref","first-page":"207","DOI":"10.1186\/bcr2607","article-title":"Breast cancer prognostic classification in the molecular era: The role of histological grade","volume":"12","author":"Rakha","year":"2010","journal-title":"Breast Cancer Res."},{"key":"ref_13","doi-asserted-by":"crossref","first-page":"457","DOI":"10.1007\/s10994-021-05946-3","article-title":"Aleatoric and epistemic uncertainty in machine learning: An introduction to concepts and methods","volume":"110","author":"Waegeman","year":"2021","journal-title":"Mach. Learn."},{"key":"ref_14","doi-asserted-by":"crossref","first-page":"130469","DOI":"10.1016\/j.eswa.2025.130469","article-title":"H&E to Oncotype DX: Predicting recurrence risk in HR+\/HER2- breast cancer","volume":"303","author":"Koyun","year":"2026","journal-title":"Expert Syst. Appl."},{"key":"ref_15","doi-asserted-by":"crossref","unstructured":"Su, Z., Niazi, M.K.K., Tavolara, T.E., Niu, S., Tozbikian, G.H., Wesolowski, R., and Gurcan, M.N. (2023). BCR-Net: A deep learning framework to predict breast cancer recurrence from histopathology images. PLoS ONE, 18.","DOI":"10.1371\/journal.pone.0283562"},{"key":"ref_16","doi-asserted-by":"crossref","first-page":"109039","DOI":"10.1016\/j.cmpb.2025.109039","article-title":"BPMambaMIL: A bio-inspired prototype-guided multiple instance learning for Oncotype DX risk assessment in histopathology","volume":"272","author":"Guo","year":"2025","journal-title":"Comput. Methods Programs Biomed."},{"key":"ref_17","doi-asserted-by":"crossref","first-page":"114","DOI":"10.1038\/s41746-025-02289-4","article-title":"Structure-aware generalization for heterogeneous histopathology via prototype-based multiple instance learning","volume":"9","author":"Yu","year":"2026","journal-title":"npj Digit. Med."},{"key":"ref_18","doi-asserted-by":"crossref","first-page":"64312","DOI":"10.52202\/079017-2051","article-title":"Leveraging tumor heterogeneity: Heterogeneous graph representation learning for cancer survival prediction in whole slide images","volume":"37","author":"Wu","year":"2024","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"ref_19","doi-asserted-by":"crossref","first-page":"btad114","DOI":"10.1093\/bioinformatics\/btad114","article-title":"Targeting tumor heterogeneity: Multiplex-detection-based multiple instance learning for whole slide image classification","volume":"39","author":"Wang","year":"2023","journal-title":"Bioinformatics"},{"key":"ref_20","doi-asserted-by":"crossref","first-page":"102885","DOI":"10.1016\/j.media.2023.102885","article-title":"An aggregation of aggregation methods in computational pathology","volume":"88","author":"Bilal","year":"2023","journal-title":"Med. Image Anal."},{"key":"ref_21","doi-asserted-by":"crossref","unstructured":"Chang, J.-R., Lee, C.-Y., Chen, C.-C., Reischl, J., Qaiser, T., and Yeh, C.-Y. (2021). Hybrid aggregation network for survival analysis from whole slide histopathological images. Medical Image Computing and Computer Assisted Intervention\u2013MICCAI 2021, Springer.","DOI":"10.1007\/978-3-030-87240-3_70"},{"key":"ref_22","unstructured":"Gallagher-Syed, A., Rossi, L., Rivellese, F., Pitzalis, C., Lewis, M., Barnes, M., and Slabaugh, G. (2023, January 20\u201324). MUSTANG: Multi-stain self-attention graph multiple instance learning pipeline for histopathology whole slide images. Proceedings of the British Machine Vision Conference, Aberdeen, UK."},{"key":"ref_23","doi-asserted-by":"crossref","unstructured":"The Cancer Genome Atlas Network (2012). Comprehensive molecular portraits of human breast tumours. Nature, 490, 61\u201370.","DOI":"10.1038\/nature11412"},{"key":"ref_24","doi-asserted-by":"crossref","first-page":"1436","DOI":"10.1016\/j.cell.2020.10.036","article-title":"Proteogenomic landscape of breast cancer tumorigenesis and targeted therapy","volume":"183","author":"Krug","year":"2020","journal-title":"Cell"},{"key":"ref_25","doi-asserted-by":"crossref","first-page":"25","DOI":"10.1038\/s41523-023-00530-5","article-title":"Integration of clinical features and deep learning on pathology for the prediction of breast cancer recurrence assays and risk of recurrence","volume":"9","author":"Howard","year":"2023","journal-title":"npj Breast Cancer"},{"key":"ref_26","doi-asserted-by":"crossref","first-page":"850","DOI":"10.1038\/s41591-024-02857-3","article-title":"Towards a general-purpose foundation model for computational pathology","volume":"30","author":"Chen","year":"2024","journal-title":"Nat. Med."},{"key":"ref_27","unstructured":"(2026, July 15). UNI Pathology Foundation Model Repository and UNI2-h Model Release. Available online: https:\/\/github.com\/mahmoodlab\/UNI."},{"key":"ref_28","unstructured":"Ilse, M., Tomczak, J., and Welling, M. (2018, January 10\u201315). Attention-based deep multiple instance learning. Proceedings of the 35th International Conference on Machine Learning, Stockholm, Sweden."},{"key":"ref_29","doi-asserted-by":"crossref","unstructured":"Bolya, D., Fu, C.-Y., Dai, X., Zhang, P., and Hoffman, J. (2022, January 23\u201327). Hydra attention: Efficient attention with many heads. Proceedings of the European Conference on Computer Vision, Tel Aviv, Israel.","DOI":"10.1007\/978-3-031-25082-8_3"},{"key":"ref_30","doi-asserted-by":"crossref","unstructured":"Zhang, H., Meng, Y., Zhao, Y., Qiao, Y., Yang, X., Coupl, S.E., and Zheng, Y. (2022, January 19\u201324). DTFT: Double-tier feature distillation multiple instance learning for histopathology whole slide image classification. Proceedings of the IEEE\/CVF Conference on Computer Vision and Pattern Recognition, New Orleans, LA, USA.","DOI":"10.1109\/CVPR52688.2022.01824"},{"key":"ref_31","doi-asserted-by":"crossref","unstructured":"Tang, W., Zhou, F., Huang, S., Zhu, X., Zhang, Y., and Liu, B. (2024, January 16\u201322). Feature re-embedding: Towards foundation model-level performance in computational pathology. Proceedings of the IEEE\/CVF Conference on Computer Vision and Pattern Recognition, Seattle, WA, USA.","DOI":"10.1109\/CVPR52733.2024.01078"},{"key":"ref_32","unstructured":"Gu, A., Goel, K., and R\u00e9, C. (2022, January 25\u201329). Efficiently modeling long sequences with structured state spaces. Proceedings of the International Conference on Learning Representations, Online."},{"key":"ref_33","first-page":"2136","article-title":"TransMIL: Transformer-based correlated multiple instance learning for whole slide image classification","volume":"34","author":"Shao","year":"2021","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"ref_34","doi-asserted-by":"crossref","first-page":"116","DOI":"10.1016\/j.breast.2019.05.006","article-title":"Nomogram update based on TAILORx clinical trial results\u2014Oncotype DX breast cancer recurrence score can be predicted using clinicopathologic data","volume":"46","author":"Orucevic","year":"2019","journal-title":"Breast"},{"key":"ref_35","unstructured":"Kaczmarzyk, J.R., Saltz, J.H., and Koo, P.K. (2024). Explainable AI for computational pathology identifies model limitations and tissue biomarkers. arXiv."}],"container-title":["Cancers"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/www.mdpi.com\/2072-6694\/18\/16\/2540\/pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,8,7]],"date-time":"2026-08-07T10:17:37Z","timestamp":1786097857000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.mdpi.com\/2072-6694\/18\/16\/2540"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,8,7]]},"references-count":35,"journal-issue":{"issue":"16","published-online":{"date-parts":[[2026,8]]}},"alternative-id":["cancers18162540"],"URL":"https:\/\/doi.org\/10.3390\/cancers18162540","relation":{},"ISSN":["2072-6694"],"issn-type":[{"value":"2072-6694","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,8,7]]}}}