{"paper":{"title":"MedSynapse-V: Bridging Visual Perception and Clinical Intuition via Latent Memory Evolution","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Medical vision-language models internalize clinical intuition by evolving latent diagnostic memories in their hidden states.","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Chunzheng Zhu, Jianxin Lin, Jiaqi Zeng, Junyu Jiang, Yijun Wang","submitted_at":"2026-04-29T04:23:35Z","abstract_excerpt":"High-precision medical diagnosis relies not only on static imaging features but also on the implicit diagnostic memory experts instantly invoke during image interpretation. We pinpoint a fundamental cognitive misalignment in medical VLMs caused by discrete tokenization, leading to quantization loss, long-range information dissipation, and missing case-adaptive expertise. To bridge this gap, we propose ours, a framework for latent diagnostic memory evolution that simulates the experiential invocation of clinicians by dynamically synthesizing implicit diagnostic memories within the model's hidde"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"ours, by transferring external expertise into endogenous parameters, significantly outperforms existing state-of-the-art methods, particularly chain-of-thought paradigms, in diagnostic accuracy.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That reinforcement learning with region-level feature masking and full-vocabulary divergence alignment can accurately quantify causal contributions of memories and internalize clinical intuition without introducing artifacts or biases.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"MedSynapse-V evolves latent diagnostic memories via meta queries, causal counterfactual refinement with RL, and dual-branch memory transition to outperform prior medical VLM methods in diagnostic accuracy.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Medical vision-language models internalize clinical intuition by evolving latent diagnostic memories in their hidden states.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"89426d6cd470e66d4f2293c894a9ddffe2eb0a3fe3729ebd2205a10698ade959"},"source":{"id":"2604.26283","kind":"arxiv","version":2},"verdict":{"id":"95b00c76-8380-41ba-bda5-112dd717ac38","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-07T13:55:37.205719Z","strongest_claim":"ours, by transferring external expertise into endogenous parameters, significantly outperforms existing state-of-the-art methods, particularly chain-of-thought paradigms, in diagnostic accuracy.","one_line_summary":"MedSynapse-V evolves latent diagnostic memories via meta queries, causal counterfactual refinement with RL, and dual-branch memory transition to outperform prior medical VLM methods in diagnostic accuracy.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That reinforcement learning with region-level feature masking and full-vocabulary divergence alignment can accurately quantify causal contributions of memories and internalize clinical intuition without introducing artifacts or biases.","pith_extraction_headline":"Medical vision-language models internalize clinical intuition by evolving latent diagnostic memories in their hidden states."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.26283/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_compliance","ran_at":"2026-05-19T20:22:38.508903Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"67d5bf5578bd1de7817b5e0c78122ff54d42c3823a50d878a8889efb37bb2a49"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","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"}