{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"institution":[{"name":"medRxiv"}],"indexed":{"date-parts":[[2026,7,13]],"date-time":"2026-07-13T09:06:21Z","timestamp":1783933581785,"version":"3.55.0"},"posted":{"date-parts":[[2026,7,9]]},"group-title":"Radiology and Imaging","reference-count":34,"publisher":"openRxiv","license":[{"start":{"date-parts":[[2026,7,9]],"date-time":"2026-07-09T00:00:00Z","timestamp":1783555200000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by-nc-nd\/4.0\/"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"accepted":{"date-parts":[[2026,7,9]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                <jats:p>Recent NCCN guideline highlighted AI-based mammographic risk prediction, but AI-based breast cancer detection remains questionable to translation. One barrier is current models often do not match routine clinical reasoning, which may add decision burden than benefits. In practice, radiologists compare current and prior mammograms while assessing breast density, bilateral symmetry, and lesion laterality. To align AI with this reasoning, we developed MuSTAF, a multi-task spatiotemporal attention fusion model for patient-level breast cancer classification from longitudinal full-field digital mammography. MuSTAF uses up to three recent mammograms, integrates temporal and cross-view information, refines suspicious-region features, and jointly predicts cancer status, breast density, and bilateral symmetry, with a separate laterality classifier for cancer-positive cases. In an internal case-control cohort (n = 351), MuSTAF achieved a cancer classification (AUC=0.84) exceeding all architecture-level baselines and published mammography AI models adapted to the same task (AUC \u2264 0.81). Simultaneously, it achieved AUCs of 0.83\/0.80 for density\/laterality assessments, and removing these auxiliary tasks reduced cancer detection performance. On the external CSAW-CC dataset (n = 8,723), model performance improved from 0.72 to 0.88 when restricting cancer cases to those with latest exams within 60 days before diagnosis, showing that temporally distant labels may shift detection evaluation toward risk prediction. Longitudinal analysis further showed that three recent exams outperformed five exams internally (AUC = 0.84 vs 0.80) and externally (0.72 vs 0.66), indicating recent imaging evidence mattered more than remote history. Overall, MuSTAF model improved longitudinal mammographic cancer classification while providing auxiliary outputs, and clarified temporal factors for applying AI to screening detection.<\/jats:p>","DOI":"10.64898\/2026.07.07.26357474","type":"posted-content","created":{"date-parts":[[2026,7,10]],"date-time":"2026-07-10T03:30:15Z","timestamp":1783654215000},"source":"Crossref","is-referenced-by-count":0,"title":["MuSTAF: Clinically Relevant Multi-task Spatiotemporal Attention Fusion Framework for Breast Cancer Detection with Longitudinal Mammography"],"prefix":"10.64898","author":[{"given":"Yutong","family":"Li","sequence":"first","affiliation":[{"name":"The University of Texas MD Anderson Cancer Center"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0009-3015-3570","authenticated-orcid":false,"given":"Austin","family":"Castelo","sequence":"additional","affiliation":[{"name":"University of Texas, MD Anderson Cancer Center"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jennifer B.","family":"Dennison","sequence":"additional","affiliation":[{"name":"The University of Texas MD Anderson Cancer Center"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Nicole M.","family":"Kettner","sequence":"additional","affiliation":[{"name":"The University of Texas MD Anderson Cancer Center"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-0085-1190","authenticated-orcid":false,"given":"Weiva","family":"Sieh","sequence":"additional","affiliation":[{"name":"University of Texas MD Anderson Cancer Center"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Joshua R.","family":"Joseph","sequence":"additional","affiliation":[{"name":"The University of Texas at Austin"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Edward","family":"Castillo","sequence":"additional","affiliation":[{"name":"The University of Texas at Austin"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Kristy","family":"Brock","sequence":"additional","affiliation":[{"name":"The University of Texas MD Anderson Cancer Center"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Olena O.","family":"Weaver","sequence":"additional","affiliation":[{"name":"The University of Texas MD Anderson Cancer Center"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Chengyue","family":"Wu","sequence":"additional","affiliation":[{"name":"The University of Texas MD Anderson Cancer Center"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"54368","reference":[{"key":"2026071301200804000_2026.07.07.26357474v1.1","unstructured":"Ferlay, J. , et al., Global Cancer Observatory: Cancer Tomorrow. Lyon, France: International Agency for Research on Cancer. Global Cancer Observatory: Cancer Tomorrow. Lyon, France: International Agency for Research on Cancer, 2020."},{"issue":"6","key":"2026071301200804000_2026.07.07.26357474v1.2","first-page":"394","volume":"68","year":"2018","journal-title":"Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA: a cancer journal for clinicians"},{"issue":"1143","key":"2026071301200804000_2026.07.07.26357474v1.3","first-page":"20220629","volume":"96","year":"2023","journal-title":"Measuring reader fatigue in the interpretation of screening digital breast tomosynthesis (DBT). The British Journal of Radiology"},{"issue":"9","key":"2026071301200804000_2026.07.07.26357474v1.4","doi-asserted-by":"crossref","first-page":"e468","DOI":"10.1016\/S2589-7500(20)30185-0","article-title":"Effect of artificial intelligence-based triaging of breast cancer screening mammograms on cancer detection and radiologist workload: a retrospective simulation study","volume":"2","year":"2020","journal-title":"The Lancet Digital Health"},{"key":"2026071301200804000_2026.07.07.26357474v1.5","doi-asserted-by":"publisher","DOI":"10.1016\/S1470-2045(23)00298-X"},{"key":"2026071301200804000_2026.07.07.26357474v1.6","doi-asserted-by":"publisher","DOI":"10.1038\/s41586-019-1799-6"},{"key":"2026071301200804000_2026.07.07.26357474v1.7","doi-asserted-by":"publisher","DOI":"10.1016\/j.breast.2022.09.007"},{"key":"2026071301200804000_2026.07.07.26357474v1.8","doi-asserted-by":"publisher","DOI":"10.1093\/jnci\/djy222"},{"issue":"4","key":"2026071301200804000_2026.07.07.26357474v1.9","first-page":"937","article-title":"Multi-task fusion for improving mammography screening data classification","volume":"41","year":"2021","journal-title":"IEEE Transactions on Medical Imaging"},{"key":"2026071301200804000_2026.07.07.26357474v1.10","doi-asserted-by":"crossref","first-page":"108101","DOI":"10.1016\/j.cmpb.2024.108101","article-title":"A multi-task fusion model based on a residual\u2013Multi-layer perceptron network for mammographic breast cancer screening","volume":"247","year":"2024","journal-title":"Computer Methods and Programs in Biomedicine"},{"key":"2026071301200804000_2026.07.07.26357474v1.11","doi-asserted-by":"crossref","unstructured":"Zhou, Z. , et al. Knowledge-Guided Multi-Task Learning for Breast Cancer Diagnosis Using Longitudinal Mammogram Images. in 2024 IEEE International Symposium on Biomedical Imaging (ISBI). 2024. IEEE.","DOI":"10.1109\/ISBI56570.2024.10635130"},{"key":"2026071301200804000_2026.07.07.26357474v1.12","doi-asserted-by":"publisher","DOI":"10.1148\/radiol.2019182716"},{"issue":"4","key":"2026071301200804000_2026.07.07.26357474v1.13","first-page":"044501","article-title":"Classifying symmetrical differences and temporal change for the detection of malignant masses in mammography using deep neural networks","volume":"4","year":"2017","journal-title":"Journal of Medical Imaging"},{"key":"2026071301200804000_2026.07.07.26357474v1.14","doi-asserted-by":"crossref","unstructured":"Zheng, Y. , C. Yang , and A. Merkulov . Breast cancer screening using convolutional neural network and follow-up digital mammography. in Computational Imaging III. 2018. SPIE.","DOI":"10.1117\/12.2304564"},{"key":"2026071301200804000_2026.07.07.26357474v1.15","doi-asserted-by":"crossref","first-page":"108919","DOI":"10.1016\/j.patcog.2022.108919","article-title":"Deep learning of longitudinal mammogram examinations for breast cancer risk prediction","volume":"132","year":"2022","journal-title":"Pattern recognition"},{"key":"2026071301200804000_2026.07.07.26357474v1.16","doi-asserted-by":"crossref","unstructured":"Wang, X. , et al. Ordinal Learning: Longitudinal Attention Alignment Model for Predicting Time to Future Breast Cancer Events from Mammograms. in International Conference on Medical Image Computing and Computer-Assisted Intervention. 2024. Springer.","DOI":"10.1007\/978-3-031-72378-0_15"},{"key":"2026071301200804000_2026.07.07.26357474v1.17","unstructured":"Strand, F. , CSAW-CC (mammography)\u2013a dataset for AI research to improve screening, diagnostics and prognostics of breast cancer (Version 1). Karolinska Institutet, 2022. 10."},{"key":"2026071301200804000_2026.07.07.26357474v1.18","doi-asserted-by":"crossref","unstructured":"Leung, J. , et al., Abstract P6-03-01: The MERIT Cohort: An MD Anderson Initiative to Integrate Blood and Imaging Biomarkers to Personalize Breast Cancer Risk. Cancer Research, 2023. 83(5_Supplement): p. P6-03-01-P6-03-01.","DOI":"10.1158\/1538-7445.SABCS22-P6-03-01"},{"key":"2026071301200804000_2026.07.07.26357474v1.19","unstructured":"Sickles, E.A. , ACR BI-RADS\u00ae Atlas Breast imaging reporting and data system. American College of Radiology., 2013: p. 39."},{"key":"2026071301200804000_2026.07.07.26357474v1.20","doi-asserted-by":"publisher","DOI":"10.1148\/radiol.2020192509"},{"issue":"1","key":"2026071301200804000_2026.07.07.26357474v1.21","first-page":"474","article-title":"Contrast limited adaptive histogram equalization","volume":"4","year":"1994","journal-title":"Graphics gems"},{"key":"2026071301200804000_2026.07.07.26357474v1.22","unstructured":"Antiga, L.P.G. , E. Stevens , and T. Viehmann , Deep learning with PyTorch. 2020: Simon and Schuster."},{"key":"2026071301200804000_2026.07.07.26357474v1.23","article-title":"Decoupled weight decay regularization","year":"2017","journal-title":"arXiv preprint"},{"key":"2026071301200804000_2026.07.07.26357474v1.24","doi-asserted-by":"crossref","unstructured":"Deng, J. , et al. Imagenet: A large-scale hierarchical image database. in 2009 IEEE conference on computer vision and pattern recognition. 2009. Ieee.","DOI":"10.1109\/CVPR.2009.5206848"},{"key":"2026071301200804000_2026.07.07.26357474v1.25","first-page":"268","article-title":"Weakly-supervised high-resolution segmentation of mammography images for breast cancer diagnosis","volume":"143","year":"2021","journal-title":"Proceedings of machine learning research"},{"key":"2026071301200804000_2026.07.07.26357474v1.26","doi-asserted-by":"crossref","unstructured":"Woo, S. , et al. Cbam: Convolutional block attention module. in Proceedings of the European conference on computer vision (ECCV). 2018.","DOI":"10.1007\/978-3-030-01234-2_1"},{"key":"2026071301200804000_2026.07.07.26357474v1.27","doi-asserted-by":"crossref","unstructured":"He, K. , et al. Deep residual learning for image recognition. in Proceedings of the IEEE conference on computer vision and pattern recognition. 2016.","DOI":"10.1109\/CVPR.2016.90"},{"key":"2026071301200804000_2026.07.07.26357474v1.28","article-title":"An image is worth 16\u00d716 words: Transformers for image recognition at scale","year":"2020","journal-title":"arXiv preprint"},{"key":"2026071301200804000_2026.07.07.26357474v1.29","doi-asserted-by":"crossref","unstructured":"Graves, A. , Long short-term memory. Supervised sequence labelling with recurrent neural networks, 2012: p. 37\u201345.","DOI":"10.1007\/978-3-642-24797-2_4"},{"key":"2026071301200804000_2026.07.07.26357474v1.30","doi-asserted-by":"publisher","DOI":"10.1162\/neco_a_01199"},{"key":"2026071301200804000_2026.07.07.26357474v1.31","doi-asserted-by":"crossref","unstructured":"Wang, C. A review on 3d convolutional neural network. in 2023 IEEE 3rd International Conference on Power, Electronics and Computer Applications (ICPECA). 2023. IEEE.","DOI":"10.1109\/ICPECA56706.2023.10075760"},{"key":"2026071301200804000_2026.07.07.26357474v1.32","doi-asserted-by":"publisher","DOI":"10.1126\/scitranslmed.aba4373"},{"key":"2026071301200804000_2026.07.07.26357474v1.33","doi-asserted-by":"crossref","first-page":"101908","DOI":"10.1016\/j.media.2020.101908","article-title":"An interpretable classifier for high-resolution breast cancer screening images utilizing weakly supervised localization","volume":"68","year":"2021","journal-title":"Medical image analysis"},{"key":"2026071301200804000_2026.07.07.26357474v1.34","doi-asserted-by":"publisher","DOI":"10.1186\/s12885-022-09484-6"}],"container-title":[],"original-title":[],"link":[{"URL":"https:\/\/syndication.highwire.org\/content\/doi\/10.64898\/2026.07.07.26357474","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,7,13]],"date-time":"2026-07-13T08:20:20Z","timestamp":1783930820000},"score":1,"resource":{"primary":{"URL":"http:\/\/medrxiv.org\/lookup\/doi\/10.64898\/2026.07.07.26357474"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,7,9]]},"references-count":34,"URL":"https:\/\/doi.org\/10.64898\/2026.07.07.26357474","relation":{},"subject":[],"published":{"date-parts":[[2026,7,9]]},"subtype":"preprint"}}