{"paper":{"title":"BrainAnytime: Anatomy-Aware Cross-Modal Pretraining for Brain Image Analysis with Arbitrary Modality Availability","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A single pretrained model analyzes brain images using whatever MRI or PET scans are available at the time.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Guangqian Yang, Qian Niu, Shujun Wang, Tong Ding, Wenlong Hou, Ye Du, Yue Xun","submitted_at":"2026-05-13T06:32:28Z","abstract_excerpt":"Clinical diagnostic workups typically follow a modality escalation pathway: after initial clinical evaluation, clinicians begin with routine structural imaging (e.g., MRI), selectively add sequences such as FLAIR or T2 to refine the differential, and reserve molecular imaging (e.g., amyloid-PET) for cases that remain uncertain after standard evaluation. Consequently, patients are observed with heterogeneous and often incomplete modality subsets. However, most current AI models assume fixed data modalities as the model inputs. In this paper, we present BrainAnytime, a unified pretraining framew"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Across four downstream tasks and five clinically motivated modality settings, BrainAnytime largely outperforms modality-specific models, missing-modality baselines, and large-scale brain MRI pretrained foundation models on most modality settings. Notably, it surpasses the strongest missing-modality baselines with relative improvements of 6.2% and 7.0% in average accuracy on CN vs. AD and CN vs. MCI classification, respectively.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The pretraining on the five chosen datasets produces representations that generalize to arbitrary unseen modality combinations and to new patient populations without retraining or fine-tuning.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A single pretrained 3D masked autoencoder handles arbitrary combinations of multi-sequence MRI and amyloid-PET for brain analysis by combining cross-modal distillation with atlas-guided curriculum masking and outperforms missing-modality baselines on Alzheimer's classification tasks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A single pretrained model analyzes brain images using whatever MRI or PET scans are available at the time.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"e2611b437b2148f21d5fc3c9b1fffb9a0064f4ff729f9c95fe8e33e53d4e4572"},"source":{"id":"2605.13059","kind":"arxiv","version":1},"verdict":{"id":"e31f57ae-9041-429f-9545-d31efd99f7ca","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:24:01.313913Z","strongest_claim":"Across four downstream tasks and five clinically motivated modality settings, BrainAnytime largely outperforms modality-specific models, missing-modality baselines, and large-scale brain MRI pretrained foundation models on most modality settings. Notably, it surpasses the strongest missing-modality baselines with relative improvements of 6.2% and 7.0% in average accuracy on CN vs. AD and CN vs. MCI classification, respectively.","one_line_summary":"A single pretrained 3D masked autoencoder handles arbitrary combinations of multi-sequence MRI and amyloid-PET for brain analysis by combining cross-modal distillation with atlas-guided curriculum masking and outperforms missing-modality baselines on Alzheimer's classification tasks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The pretraining on the five chosen datasets produces representations that generalize to arbitrary unseen modality combinations and to new patient populations without retraining or fine-tuning.","pith_extraction_headline":"A single pretrained model analyzes brain images using whatever MRI or PET scans are available at the time."},"references":{"count":27,"sample":[{"doi":"","year":2007,"title":"Alzheimer Disease & Associated Disorders21, 249–258 (2007)","work_id":"8a22d7d4-2668-42b3-a76b-0a55f7a769a5","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2006,"title":"Acta Neuropathologica112, 389 – 404 (2006)","work_id":"03bd51c1-55fc-4655-bd65-da6ca719e374","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Alzheimer’s & Dementia21(2024)","work_id":"5824dee4-6848-4980-af0c-a7c52cee37de","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"The Lancet Neurology19(11), 951–962 (2020)","work_id":"c7596758-a0b9-4b0d-8447-bd4a43c25275","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Pattern Recognition p","work_id":"2e2a401a-43fd-4725-80a3-46c7fa4b70ea","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":27,"snapshot_sha256":"1e69089fceccf84b84fb871d8f6e7debd35a18077d506fc43bc92999b9a042de","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}