{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"institution":[{"name":"bioRxiv"}],"indexed":{"date-parts":[[2026,5,16]],"date-time":"2026-05-16T02:57:18Z","timestamp":1778900238109,"version":"3.51.4"},"posted":{"date-parts":[[2026,1,2]]},"group-title":"Cell Biology","reference-count":46,"publisher":"openRxiv","license":[{"start":{"date-parts":[[2026,1,2]],"date-time":"2026-01-02T00:00:00Z","timestamp":1767312000000},"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":[[2026,1,25]]},"abstract":"<jats:title>ABSTRACT<\/jats:title>\n                <jats:p>Quantitative, time-resolved 3D fluorescence microscopy can reveal complex cellular dynamics in living cells and tissues. Broader use remains limited by the difficulty of identifying, segmenting, and tracking objects of different size and shape in crowded intracellular environments in low-contrast, anisotropic, monochromatic image volumes. Objects overlap, deform, appear and disappear, and span wide ranges of size and intensity. Classical segmentation pipelines typically require high signal-to-noise data and rely on intensity heuristics with hand-tuned postprocessing that generalize poorly. Supervised deep learning methods require extensive voxel-level annotations that are costly, inconsistent across phenotypes, and rapidly become obsolete as imaging conditions change.<\/jats:p>\n                <jats:p>We introduce SpatialDINO, a fully automated self-supervised method that trains a native 3D vision transformer, based on a modified version of DINOv2 (1). SpatialDINO yields robust semantic feature maps from single channels of multi-channel microscopy that, irrespective of object shape, support object detection and segmentation directly from na\u00efve 3D images across z-spacings and numbers of planes and different imaging modalities, without retraining or voxel annotations.<\/jats:p>\n                <jats:p>We trained SpatialDINO on a small set of confocal volumes acquired by live-cell fluorescent 3D lattice light-sheet microscopy, spanning targets of different size and shape located in crowded cellular environments, from diffraction-limited clathrin coated pits and clathrin coated vesicles to bigger structures including endosomes and lysosomes, and endosomes and lysosomes pharmacologically enlarged to highlight endosomal membrane profiles. Post-processing of the features generated by SpatialDINO allows detection and unique object identification of these objects in na\u00efve 3D images. It also enables detection of significantly different previously unseen object classes, such as cellular plasma membranes and nuclei and even tumors in MRI scans. Finally, we illustrate its value by tracking endosomes in 3D time series, combining SpatialDINO-derived feature similarity with spatial proximity to improve association through occlusion, abrupt appearance changes, and dense packing, all conditions that have been challenging for existing methods.<\/jats:p>\n                <jats:p>SpatialDINO therefore lowers a major barrier to quantitative analysis of heterogeneous, monochromatic objects in crowded 3D cellular environments.<\/jats:p>\n                <jats:sec>\n                  <jats:title>SUMMARY<\/jats:title>\n                  <jats:p>Lavaee et al. have developed SpatialDINO to surmount the difficulties presented by attempting to identify, segment, and track objects in crowded volumes. SpatialDINO is a self-supervised, native 3D vision transformer trained directly on unlabeled fluorescence volumes acquired by live-cell 3D lattice light-sheet microscopy. By learning dense volumetric representations without voxel-level supervision, SpatialDINO generates features that enable fully automated detection, segmentation, and tracking of subcellular structures across a wide range of sizes and morphologies in crowded, anisotropic 3D\/4D datasets acquired with different microscopy modalities. The approach generalizes across targets and imaging conditions without further training, reducing dependence on manual annotation while maintaining performance in complex cellular environments.<\/jats:p>\n                <\/jats:sec>\n                <jats:sec>\n                  <jats:title>SIGNIFICANCE<\/jats:title>\n                  <jats:p>SpatialDINO brings a self-supervised foundation model for analyzing 3D fluorescence microscopy images by adapting DINOv2-style joint-embedding training to learn dense volumetric features directly from unlabeled 3D datasets. By exploiting true 3D context rather than slice-wise \u201c2.5D\u201d aggregation, it enables automated detection and segmentation in crowded, anisotropic, low-contrast volume and enables tracking in 4D time-lapse data. SpatialDINO generalizes across targets and imaging conditions without voxel-level annotation or retraining.<\/jats:p>\n                <\/jats:sec>","DOI":"10.64898\/2025.12.31.697247","type":"posted-content","created":{"date-parts":[[2026,1,2]],"date-time":"2026-01-02T22:15:13Z","timestamp":1767392113000},"source":"Crossref","is-referenced-by-count":1,"title":["SpatialDINO: A Self-Supervised 3D Vision Transformer that enables Segmentation and Tracking in Crowded Cellular Environments"],"prefix":"10.64898","author":[{"ORCID":"https:\/\/orcid.org\/0009-0006-0751-193X","authenticated-orcid":false,"given":"Alex","family":"Lavaee","sequence":"first","affiliation":[{"name":"Program in Cellular and Molecular Medicine, Boston Children\u2019s Hospital, 200 Longwood Ave, Boston, MA 02115, USA"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-2692-7472","authenticated-orcid":false,"given":"Arkash","family":"Jain","sequence":"additional","affiliation":[{"name":"Program in Cellular and Molecular Medicine, Boston Children\u2019s Hospital, 200 Longwood Ave, Boston, MA 02115, USA"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-4001-3389","authenticated-orcid":false,"given":"Gustavo","family":"Scanavachi","sequence":"additional","affiliation":[{"name":"Program in Cellular and Molecular Medicine, Boston Children\u2019s Hospital, 200 Longwood Ave, Boston, MA 02115, USA"},{"name":"Department of Cell Biology, Harvard Medical School, 200 Longwood Ave, Boston, MA 02115, USA"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-3531-0987","authenticated-orcid":false,"given":"Jose Inacio","family":"Costa-Filho","sequence":"additional","affiliation":[{"name":"Program in Cellular and Molecular Medicine, Boston Children\u2019s Hospital, 200 Longwood Ave, Boston, MA 02115, USA"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Adam","family":"Ingemansson","sequence":"additional","affiliation":[{"name":"Program in Cellular and Molecular Medicine, Boston Children\u2019s Hospital, 200 Longwood Ave, Boston, MA 02115, USA"},{"name":"Royal Institute of Technology, Stockholm, Sweden"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-0559-893X","authenticated-orcid":false,"given":"Tom","family":"Kirchhausen","sequence":"additional","affiliation":[{"name":"Program in Cellular and Molecular Medicine, Boston Children\u2019s Hospital, 200 Longwood Ave, Boston, MA 02115, USA"},{"name":"Department of Cell Biology, Harvard Medical School, 200 Longwood Ave, Boston, MA 02115, USA"},{"name":"Department of Pediatrics, Harvard Medical School, 200 Longwood Ave, Boston, MA 02115, USA"}],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"54368","reference":[{"key":"2026020502142850000_2025.12.31.697247v2.1","doi-asserted-by":"publisher","DOI":"10.48550\/arxiv.2304.07193"},{"key":"2026020502142850000_2025.12.31.697247v2.2","doi-asserted-by":"publisher","DOI":"10.1038\/s41592-024-02580-4"},{"key":"2026020502142850000_2025.12.31.697247v2.3","doi-asserted-by":"publisher","DOI":"10.48550\/arxiv.2104.14294"},{"key":"2026020502142850000_2025.12.31.697247v2.4","doi-asserted-by":"publisher","DOI":"10.1101\/2023.06.16.545359"},{"key":"2026020502142850000_2025.12.31.697247v2.5","doi-asserted-by":"publisher","DOI":"10.48550\/arxiv.2405.15700"},{"key":"2026020502142850000_2025.12.31.697247v2.6","doi-asserted-by":"crossref","first-page":"e1013071","DOI":"10.1371\/journal.pcbi.1013071","article-title":"Cell-TRACTR: A transformer-based model for end-to-end segmentation and tracking of cells","volume":"21","year":"2025","journal-title":"PLOS Comput. Biol"},{"key":"2026020502142850000_2025.12.31.697247v2.7","doi-asserted-by":"publisher","DOI":"10.48550\/arxiv.2509.12512"},{"key":"2026020502142850000_2025.12.31.697247v2.8","doi-asserted-by":"publisher","DOI":"10.48550\/arxiv.2312.02366"},{"key":"2026020502142850000_2025.12.31.697247v2.9","doi-asserted-by":"crossref","first-page":"23979","DOI":"10.1038\/s41598-025-09041-8","article-title":"Medical slice transformer for improved diagnosis and explainability on 3D medical images with DINOv2","volume":"15","year":"2025","journal-title":"Sci. Rep"},{"key":"2026020502142850000_2025.12.31.697247v2.10","doi-asserted-by":"crossref","first-page":"52","DOI":"10.1186\/s12938-024-01238-8","article-title":"Towards more precise automatic analysis: a systematic review of deep learning-based multi-organ segmentation","volume":"23","year":"2024","journal-title":"Biomed. Eng. OnLine"},{"key":"2026020502142850000_2025.12.31.697247v2.11","doi-asserted-by":"publisher","DOI":"10.1073\/pnas.2007837117"},{"key":"2026020502142850000_2025.12.31.697247v2.12","doi-asserted-by":"publisher","DOI":"10.1016\/j.devcel.2021.05.015"},{"key":"2026020502142850000_2025.12.31.697247v2.13","doi-asserted-by":"publisher","DOI":"10.1016\/j.devcel.2024.03.021"},{"key":"2026020502142850000_2025.12.31.697247v2.14","doi-asserted-by":"publisher","DOI":"10.1038\/s41592-019-0539-z"},{"key":"2026020502142850000_2025.12.31.697247v2.15","doi-asserted-by":"publisher","DOI":"10.1101\/2023.04.12.536640"},{"key":"2026020502142850000_2025.12.31.697247v2.16","doi-asserted-by":"publisher","DOI":"10.1101\/2025.03.09.642092"},{"key":"2026020502142850000_2025.12.31.697247v2.17","doi-asserted-by":"publisher","DOI":"10.48550\/arxiv.2203.12602"},{"key":"2026020502142850000_2025.12.31.697247v2.18","unstructured":"A. Kazemnejad , I. Padhi , K. Natesan , P. Das , S. Reddy , The Impact of Positional Encoding on Length Generalization in Transformers in Thirty-Seventh Conference on Neural Information Processing Systems, (2023)."},{"key":"2026020502142850000_2025.12.31.697247v2.19","doi-asserted-by":"crossref","unstructured":"Haviv, et al., Transformer Language Models without Positional Encodings Still Learn Positional Information in Findings of the Association for Computational Linguistics: EMNLP 2022, (Association for Computational Linguistics, 2022), pp. 1382\u20131390.","DOI":"10.18653\/v1\/2022.findings-emnlp.99"},{"key":"2026020502142850000_2025.12.31.697247v2.20","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-031-73013-9_26"},{"key":"2026020502142850000_2025.12.31.697247v2.21","doi-asserted-by":"publisher","DOI":"10.1091\/mbc.e16-03-0164"},{"key":"2026020502142850000_2025.12.31.697247v2.22","doi-asserted-by":"publisher","DOI":"10.1038\/s41592-022-01507-1"},{"key":"2026020502142850000_2025.12.31.697247v2.23","doi-asserted-by":"publisher","DOI":"10.1016\/j.devcel.2013.06.019"},{"key":"2026020502142850000_2025.12.31.697247v2.24","doi-asserted-by":"crossref","first-page":"695","DOI":"10.1038\/nmeth.1237","article-title":"Robust single-particle tracking in live-cell time-lapse sequences","volume":"5","year":"2008","journal-title":"Nature Methods"},{"key":"2026020502142850000_2025.12.31.697247v2.25","unstructured":"A. Paszke , et al., PyTorch: an imperative style, high-performance deep learning library. (2019)."},{"key":"2026020502142850000_2025.12.31.697247v2.26","first-page":"4037","article-title":"Self-Supervised Visual Feature Learning With Deep Neural Networks: A Survey","volume":"43","year":"2019","journal-title":"IEEE Trans. Pattern Anal. Mach. Intell"},{"key":"2026020502142850000_2025.12.31.697247v2.27","unstructured":"R. Balestriero , Y. LeCun , How learning by reconstruction produces uninformative features for perception in Proceedings of the 41st International Conference on Machine Learning, ICML\u201924., (JMLR.org, 2024)."},{"key":"2026020502142850000_2025.12.31.697247v2.28","doi-asserted-by":"publisher","DOI":"10.48550\/arxiv.2505.12477"},{"key":"2026020502142850000_2025.12.31.697247v2.29","unstructured":"Z. Wu , et al., Simplifying DINO via Coding Rate Regularization in Forty-Second International Conference on Machine Learning, (2025)."},{"key":"2026020502142850000_2025.12.31.697247v2.30","unstructured":"M. Oquab , et al., DINOv2: Learning Robust Visual Features without Supervision. Transactions on Machine Learning Research (2024)."},{"key":"2026020502142850000_2025.12.31.697247v2.31","unstructured":"H. Govindarajan , P. Sid\u00c3\u00a9n , J. Roll , F. Lindsten , On Partial Prototype Collapse in the DINO Family of Self-Supervised Methods in 35th British Machine Vision Conference 2024, BMVC 2024, Glasgow, UK, November 25-28, 2024, (BMVA, 2024)."},{"key":"2026020502142850000_2025.12.31.697247v2.32","unstructured":"A. Vaswani , et al., Attention is All you Need in Advances in Neural Information Processing Systems, (Curran Associates, Inc., 2017)."},{"key":"2026020502142850000_2025.12.31.697247v2.33","unstructured":"A. Dosovitskiy , et al., An Image is Worth 16\u00d716 Words: Transformers for Image Recognition at Scale in International Conference on Learning Representations, (2021)."},{"key":"2026020502142850000_2025.12.31.697247v2.34","unstructured":"T. Darcet , M. Oquab , J. Mairal , P. Bojanowski , Vision Transformers Need Registers in The Twelfth International Conference on Learning Representations, (2024)."},{"key":"2026020502142850000_2025.12.31.697247v2.35","first-page":"9630","article-title":"Emerging Properties in Self-Supervised Vision Transformers","volume":"00","year":"2021","journal-title":"IEEECVF Int. Conf. Comput. Vis. (ICCV)"},{"key":"2026020502142850000_2025.12.31.697247v2.36","unstructured":"J. Zhou , et al., Image BERT Pre-training with Online Tokenizer in International Conference on Learning Representations, (2022)."},{"key":"2026020502142850000_2025.12.31.697247v2.37","doi-asserted-by":"publisher","DOI":"10.48550\/arxiv.2302.04303"},{"key":"2026020502142850000_2025.12.31.697247v2.38","doi-asserted-by":"publisher","DOI":"10.48550\/arxiv.2506.08010"},{"key":"2026020502142850000_2025.12.31.697247v2.39","doi-asserted-by":"publisher","DOI":"10.48550\/arxiv.1905.12322"},{"key":"2026020502142850000_2025.12.31.697247v2.40","doi-asserted-by":"publisher","DOI":"10.14778\/3415478.3415530"},{"key":"2026020502142850000_2025.12.31.697247v2.41","doi-asserted-by":"crossref","first-page":"3848","DOI":"10.14778\/3611540.3611569","article-title":"PyTorch FSDP: Experiences on Scaling Fully Sharded Data Parallel","volume":"16","year":"2023","journal-title":"Proc. VLDB Endow"},{"key":"2026020502142850000_2025.12.31.697247v2.42","doi-asserted-by":"publisher","DOI":"10.1109\/cvpr.2009.5206848"},{"key":"2026020502142850000_2025.12.31.697247v2.43","doi-asserted-by":"crossref","unstructured":"K. Zuiderveld , Contrast limited adaptive histogram equalization. 474\u2013485 (1994).","DOI":"10.1016\/B978-0-12-336156-1.50061-6"},{"key":"2026020502142850000_2025.12.31.697247v2.44","first-page":"15979","article-title":"Masked Autoencoders Are Scalable Vision Learners","volume":"00","year":"2022","journal-title":"IEEECVF Conf. Comput. Vis. Pattern Recognit. (CVPR)"},{"key":"2026020502142850000_2025.12.31.697247v2.45","doi-asserted-by":"crossref","first-page":"2386","DOI":"10.1038\/s41592-025-02887-w","article-title":"Universal consensus 3D segmentation of cells from 2D segmented stacks","volume":"22","year":"2025","journal-title":"Nat. Methods"},{"key":"2026020502142850000_2025.12.31.697247v2.46","doi-asserted-by":"publisher","DOI":"10.1091\/mbc.E11-08-0731"}],"container-title":[],"original-title":[],"link":[{"URL":"https:\/\/syndication.highwire.org\/content\/doi\/10.64898\/2025.12.31.697247","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,2,5]],"date-time":"2026-02-05T10:14:45Z","timestamp":1770286485000},"score":1,"resource":{"primary":{"URL":"http:\/\/biorxiv.org\/lookup\/doi\/10.64898\/2025.12.31.697247"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,1,2]]},"references-count":46,"URL":"https:\/\/doi.org\/10.64898\/2025.12.31.697247","relation":{},"subject":[],"published":{"date-parts":[[2026,1,2]]},"subtype":"preprint"}}