{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:DZFN4MODJQ7FNBLFAGOKFDHD4B","short_pith_number":"pith:DZFN4MOD","canonical_record":{"source":{"id":"2605.12650","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-12T18:56:34Z","cross_cats_sorted":[],"title_canon_sha256":"1fbd74dc3032483fcb42cb4bf868c496674f845e0b5dd39bcec3c0fb86990279","abstract_canon_sha256":"6f89acd400b7c6983a59a22be47d3b707f6362208b685f5822a2fbb7aad13963"},"schema_version":"1.0"},"canonical_sha256":"1e4ade31c34c3e568565019ca28ce3e0695785162ef84b12762d910c106ed973","source":{"kind":"arxiv","id":"2605.12650","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.12650","created_at":"2026-05-18T03:09:59Z"},{"alias_kind":"arxiv_version","alias_value":"2605.12650v1","created_at":"2026-05-18T03:09:59Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.12650","created_at":"2026-05-18T03:09:59Z"},{"alias_kind":"pith_short_12","alias_value":"DZFN4MODJQ7F","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"DZFN4MODJQ7FNBLF","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"DZFN4MOD","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:DZFN4MODJQ7FNBLFAGOKFDHD4B","target":"record","payload":{"canonical_record":{"source":{"id":"2605.12650","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-12T18:56:34Z","cross_cats_sorted":[],"title_canon_sha256":"1fbd74dc3032483fcb42cb4bf868c496674f845e0b5dd39bcec3c0fb86990279","abstract_canon_sha256":"6f89acd400b7c6983a59a22be47d3b707f6362208b685f5822a2fbb7aad13963"},"schema_version":"1.0"},"canonical_sha256":"1e4ade31c34c3e568565019ca28ce3e0695785162ef84b12762d910c106ed973","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T03:09:59.747517Z","signature_b64":"MTqRG/9NSzH+vxMUZebZw1u5hz2/Yc4Wk8fbectjD20R9ddE4KZZSOkG07u/BuXWudAzLVRfu1BZ1By1jaNrDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1e4ade31c34c3e568565019ca28ce3e0695785162ef84b12762d910c106ed973","last_reissued_at":"2026-05-18T03:09:59.746996Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T03:09:59.746996Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.12650","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T03:09:59Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"OH+j2tZCyk10NvN/aakoZV5PCb2V/38cMKuNvXX9O2Ey7RMNUhpmUNuyaVmaLj8cSTqkTD6/mEETOyBj1EmUCw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-01T17:48:28.922300Z"},"content_sha256":"d864ab9f3b947168753f629febdb004622e18614b30228db522929162ca56a8d","schema_version":"1.0","event_id":"sha256:d864ab9f3b947168753f629febdb004622e18614b30228db522929162ca56a8d"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:DZFN4MODJQ7FNBLFAGOKFDHD4B","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"CRAFT: Clinical Reward-Aligned Finetuning for Medical Image Synthesis","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Clinical reward finetuning lets diffusion models generate medical images that better match pathology criteria and improve downstream classifiers.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Alex El Darzi, Carlo El Khoury, Han Feng, Jihun Hamm, Nassir Marrouche, Yunsung Chung","submitted_at":"2026-05-12T18:56:34Z","abstract_excerpt":"Foundation diffusion models can generate photorealistic natural images, but adapting them to medical imaging remains challenging. In medical adaptation, limited labeled data can exacerbate hallucination-like and clinically implausible synthesis, while existing metrics such as FID or Inception Score do not quantify per-image alignment with pathology-relevant criteria. We introduce the Clinical Alignment Score (CAS), a foundation-model-based proxy for clinical alignment that evaluates generated images along four complementary dimensions beyond visual fidelity. Building on CAS, we propose Clinica"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Across four diverse modalities, CRAFT improves CAS and downstream classification performance over strong adaptation baselines. Beyond average CAS gains, CRAFT reduces the empirical low-alignment tail below a real-image reference threshold by 5.5-34.7% points relative to the strongest baseline, corresponding to a 20.4% average relative reduction across datasets.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the newly introduced Clinical Alignment Score (CAS) serves as a reliable proxy for actual clinical plausibility and pathology relevance, and that optimizing the diffusion model against rewards derived from it produces genuinely improved medical images rather than artifacts tuned to the proxy.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"CRAFT adapts diffusion models to medical images via clinical reward alignment from LLMs and VLMs, improving alignment scores and cutting low-quality generations by 20.4% on average across modalities.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Clinical reward finetuning lets diffusion models generate medical images that better match pathology criteria and improve downstream classifiers.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"e2dd463c9c41d82932626f50f48199f16db99d136ceb73a8bb0b6a443715f55b"},"source":{"id":"2605.12650","kind":"arxiv","version":1},"verdict":{"id":"ceaa3b2e-058a-4ff7-a2b0-af1445ea181a","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T21:06:21.195838Z","strongest_claim":"Across four diverse modalities, CRAFT improves CAS and downstream classification performance over strong adaptation baselines. Beyond average CAS gains, CRAFT reduces the empirical low-alignment tail below a real-image reference threshold by 5.5-34.7% points relative to the strongest baseline, corresponding to a 20.4% average relative reduction across datasets.","one_line_summary":"CRAFT adapts diffusion models to medical images via clinical reward alignment from LLMs and VLMs, improving alignment scores and cutting low-quality generations by 20.4% on average across modalities.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the newly introduced Clinical Alignment Score (CAS) serves as a reliable proxy for actual clinical plausibility and pathology relevance, and that optimizing the diffusion model against rewards derived from it produces genuinely improved medical images rather than artifacts tuned to the proxy.","pith_extraction_headline":"Clinical reward finetuning lets diffusion models generate medical images that better match pathology criteria and improve downstream classifiers."},"references":{"count":22,"sample":[{"doi":"","year":null,"title":"GPT-4 Technical Report","work_id":"b928e041-6991-4c08-8c81-0359e4097c7b","ref_index":1,"cited_arxiv_id":"2303.08774","is_internal_anchor":true},{"doi":"","year":null,"title":"Towards bet- ter optimization for listwise preference in diffusion models","work_id":"3424a812-ad0b-41e6-814d-2d4d9c9702b4","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Demystifying MMD GANs","work_id":"92565b9d-3665-4c64-ab7c-3451cfd6796c","ref_index":3,"cited_arxiv_id":"1801.01401","is_internal_anchor":true},{"doi":"","year":null,"title":"Training Diffusion Models with Reinforcement Learning","work_id":"67684dda-3930-452a-b91a-36cbb8e2e219","ref_index":4,"cited_arxiv_id":"2305.13301","is_internal_anchor":true},{"doi":"","year":null,"title":"Meta clip 2: A worldwide scaling recipe","work_id":"35c3d2a3-1b34-4131-b73a-976f4727f77f","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":22,"snapshot_sha256":"c59bc3939dfc19f9aa73382c89c3a32fe40feaa475b277b7b8ddc073951d9e3f","internal_anchors":10},"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"},"verdict_id":"ceaa3b2e-058a-4ff7-a2b0-af1445ea181a"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T03:09:59Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"/eZFB2dugmDEJ/5AVPLTt9ABUq8VP/XRXq7WhBCF6axiyDph4wgYVYY1G+7mMHiCS66FJAoaFICiIxZHOCYCCQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-01T17:48:28.922869Z"},"content_sha256":"f87a51947860f958d0a8579464d0a8884069364040228e00e2691a5b8e6e070a","schema_version":"1.0","event_id":"sha256:f87a51947860f958d0a8579464d0a8884069364040228e00e2691a5b8e6e070a"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/DZFN4MODJQ7FNBLFAGOKFDHD4B/bundle.json","state_url":"https://pith.science/pith/DZFN4MODJQ7FNBLFAGOKFDHD4B/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/DZFN4MODJQ7FNBLFAGOKFDHD4B/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-06-01T17:48:28Z","links":{"resolver":"https://pith.science/pith/DZFN4MODJQ7FNBLFAGOKFDHD4B","bundle":"https://pith.science/pith/DZFN4MODJQ7FNBLFAGOKFDHD4B/bundle.json","state":"https://pith.science/pith/DZFN4MODJQ7FNBLFAGOKFDHD4B/state.json","well_known_bundle":"https://pith.science/.well-known/pith/DZFN4MODJQ7FNBLFAGOKFDHD4B/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:DZFN4MODJQ7FNBLFAGOKFDHD4B","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"6f89acd400b7c6983a59a22be47d3b707f6362208b685f5822a2fbb7aad13963","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-12T18:56:34Z","title_canon_sha256":"1fbd74dc3032483fcb42cb4bf868c496674f845e0b5dd39bcec3c0fb86990279"},"schema_version":"1.0","source":{"id":"2605.12650","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.12650","created_at":"2026-05-18T03:09:59Z"},{"alias_kind":"arxiv_version","alias_value":"2605.12650v1","created_at":"2026-05-18T03:09:59Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.12650","created_at":"2026-05-18T03:09:59Z"},{"alias_kind":"pith_short_12","alias_value":"DZFN4MODJQ7F","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"DZFN4MODJQ7FNBLF","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"DZFN4MOD","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:f87a51947860f958d0a8579464d0a8884069364040228e00e2691a5b8e6e070a","target":"graph","created_at":"2026-05-18T03:09:59Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Across four diverse modalities, CRAFT improves CAS and downstream classification performance over strong adaptation baselines. Beyond average CAS gains, CRAFT reduces the empirical low-alignment tail below a real-image reference threshold by 5.5-34.7% points relative to the strongest baseline, corresponding to a 20.4% average relative reduction across datasets."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the newly introduced Clinical Alignment Score (CAS) serves as a reliable proxy for actual clinical plausibility and pathology relevance, and that optimizing the diffusion model against rewards derived from it produces genuinely improved medical images rather than artifacts tuned to the proxy."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"CRAFT adapts diffusion models to medical images via clinical reward alignment from LLMs and VLMs, improving alignment scores and cutting low-quality generations by 20.4% on average across modalities."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Clinical reward finetuning lets diffusion models generate medical images that better match pathology criteria and improve downstream classifiers."}],"snapshot_sha256":"e2dd463c9c41d82932626f50f48199f16db99d136ceb73a8bb0b6a443715f55b"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Foundation diffusion models can generate photorealistic natural images, but adapting them to medical imaging remains challenging. In medical adaptation, limited labeled data can exacerbate hallucination-like and clinically implausible synthesis, while existing metrics such as FID or Inception Score do not quantify per-image alignment with pathology-relevant criteria. We introduce the Clinical Alignment Score (CAS), a foundation-model-based proxy for clinical alignment that evaluates generated images along four complementary dimensions beyond visual fidelity. Building on CAS, we propose Clinica","authors_text":"Alex El Darzi, Carlo El Khoury, Han Feng, Jihun Hamm, Nassir Marrouche, Yunsung Chung","cross_cats":[],"headline":"Clinical reward finetuning lets diffusion models generate medical images that better match pathology criteria and improve downstream classifiers.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-12T18:56:34Z","title":"CRAFT: Clinical Reward-Aligned Finetuning for Medical Image Synthesis"},"references":{"count":22,"internal_anchors":10,"resolved_work":22,"sample":[{"cited_arxiv_id":"2303.08774","doi":"","is_internal_anchor":true,"ref_index":1,"title":"GPT-4 Technical Report","work_id":"b928e041-6991-4c08-8c81-0359e4097c7b","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Towards bet- ter optimization for listwise preference in diffusion models","work_id":"3424a812-ad0b-41e6-814d-2d4d9c9702b4","year":null},{"cited_arxiv_id":"1801.01401","doi":"","is_internal_anchor":true,"ref_index":3,"title":"Demystifying MMD GANs","work_id":"92565b9d-3665-4c64-ab7c-3451cfd6796c","year":null},{"cited_arxiv_id":"2305.13301","doi":"","is_internal_anchor":true,"ref_index":4,"title":"Training Diffusion Models with Reinforcement Learning","work_id":"67684dda-3930-452a-b91a-36cbb8e2e219","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Meta clip 2: A worldwide scaling recipe","work_id":"35c3d2a3-1b34-4131-b73a-976f4727f77f","year":null}],"snapshot_sha256":"c59bc3939dfc19f9aa73382c89c3a32fe40feaa475b277b7b8ddc073951d9e3f"},"source":{"id":"2605.12650","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T21:06:21.195838Z","id":"ceaa3b2e-058a-4ff7-a2b0-af1445ea181a","model_set":{"reader":"grok-4.3"},"one_line_summary":"CRAFT adapts diffusion models to medical images via clinical reward alignment from LLMs and VLMs, improving alignment scores and cutting low-quality generations by 20.4% on average across modalities.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Clinical reward finetuning lets diffusion models generate medical images that better match pathology criteria and improve downstream classifiers.","strongest_claim":"Across four diverse modalities, CRAFT improves CAS and downstream classification performance over strong adaptation baselines. Beyond average CAS gains, CRAFT reduces the empirical low-alignment tail below a real-image reference threshold by 5.5-34.7% points relative to the strongest baseline, corresponding to a 20.4% average relative reduction across datasets.","weakest_assumption":"That the newly introduced Clinical Alignment Score (CAS) serves as a reliable proxy for actual clinical plausibility and pathology relevance, and that optimizing the diffusion model against rewards derived from it produces genuinely improved medical images rather than artifacts tuned to the proxy."}},"verdict_id":"ceaa3b2e-058a-4ff7-a2b0-af1445ea181a"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:d864ab9f3b947168753f629febdb004622e18614b30228db522929162ca56a8d","target":"record","created_at":"2026-05-18T03:09:59Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"6f89acd400b7c6983a59a22be47d3b707f6362208b685f5822a2fbb7aad13963","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-12T18:56:34Z","title_canon_sha256":"1fbd74dc3032483fcb42cb4bf868c496674f845e0b5dd39bcec3c0fb86990279"},"schema_version":"1.0","source":{"id":"2605.12650","kind":"arxiv","version":1}},"canonical_sha256":"1e4ade31c34c3e568565019ca28ce3e0695785162ef84b12762d910c106ed973","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"1e4ade31c34c3e568565019ca28ce3e0695785162ef84b12762d910c106ed973","first_computed_at":"2026-05-18T03:09:59.746996Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T03:09:59.746996Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"MTqRG/9NSzH+vxMUZebZw1u5hz2/Yc4Wk8fbectjD20R9ddE4KZZSOkG07u/BuXWudAzLVRfu1BZ1By1jaNrDg==","signature_status":"signed_v1","signed_at":"2026-05-18T03:09:59.747517Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.12650","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:d864ab9f3b947168753f629febdb004622e18614b30228db522929162ca56a8d","sha256:f87a51947860f958d0a8579464d0a8884069364040228e00e2691a5b8e6e070a"],"state_sha256":"4e5074b69a13f667565c0911f6c73d0565084573c662bf6f2ce0fab4008c9afe"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"CE7MEy9O+qyYq+F4oGBWKjNDu8disLRcOeSxmWYdBTR1EHXpGoUxHOohbxgNmRVVtKgzJEe9aj6/a52srlCMDw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-01T17:48:28.926040Z","bundle_sha256":"aa6361a69d241209d4e24b844b9c6c471964e40d6aff3fa6c591bf3654042d8f"}}