{"paper":{"title":"MedCore: Boundary-Preserving Medical Core Pruning for MedSAM","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"MedCore prunes MedSAM parameters by 60 percent while preserving boundary fidelity through dual-intervention scoring and logit-level boundary leverage.","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Cenwei Zhang, Lei You, Suncheng Xiang","submitted_at":"2026-05-13T15:42:39Z","abstract_excerpt":"Medical segmentation foundation models such as SAM and MedSAM provide strong prompt-driven segmentation, but their image encoders are still too large for many clinical settings. Compression is also risky in medicine because a model can keep high Dice while losing boundary fidelity. We propose MedCore, a structured pruning framework for MedSAM. The main idea is to preserve two kinds of structures: structures that became important during SAM-to-MedSAM adaptation, and structures that have high boundary leverage. We identify the first type by a dual-intervention score that compares zeroing a group"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"On polyp segmentation benchmarks, MedCore reduces parameters by 60.0% and FLOPs by 58.4% while achieving Dice 0.9549, Boundary F1 0.6388, and HD95 5.14 after recovery fine-tuning. It also reaches 86.6% parameter reduction with strong boundary quality.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the dual-intervention score and boundary-aware Fisher estimation correctly identify preservable structures without hidden degradation, and that the boundary leverage principle accurately predicts and controls compression-induced boundary displacement in practice.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"MedCore achieves 60% parameter and 58.4% FLOP reduction on MedSAM with Dice 0.9549 and preserved boundary metrics via dual-intervention pruning and a new boundary leverage principle.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"MedCore prunes MedSAM parameters by 60 percent while preserving boundary fidelity through dual-intervention scoring and logit-level boundary leverage.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"ffdf49ef323b6afecf8789f0d18464be073ce307234611eb26e523f00fbb9fda"},"source":{"id":"2605.13688","kind":"arxiv","version":1},"verdict":{"id":"c52ce991-7801-4513-b82a-cf5192dfed92","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:09:43.803151Z","strongest_claim":"On polyp segmentation benchmarks, MedCore reduces parameters by 60.0% and FLOPs by 58.4% while achieving Dice 0.9549, Boundary F1 0.6388, and HD95 5.14 after recovery fine-tuning. It also reaches 86.6% parameter reduction with strong boundary quality.","one_line_summary":"MedCore achieves 60% parameter and 58.4% FLOP reduction on MedSAM with Dice 0.9549 and preserved boundary metrics via dual-intervention pruning and a new boundary leverage principle.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the dual-intervention score and boundary-aware Fisher estimation correctly identify preservable structures without hidden degradation, and that the boundary leverage principle accurately predicts and controls compression-induced boundary displacement in practice.","pith_extraction_headline":"MedCore prunes MedSAM parameters by 60 percent while preserving boundary fidelity through dual-intervention scoring and logit-level boundary leverage."},"references":{"count":51,"sample":[{"doi":"","year":2023,"title":"Segment Anything","work_id":"2bbf46ca-720a-45a1-8e9c-10c33fbeada0","ref_index":1,"cited_arxiv_id":"2304.02643","is_internal_anchor":true},{"doi":"","year":2024,"title":"Segment anything in medical images.Nature Communications, 15:654, 2024","work_id":"27d74308-606e-450f-9c86-8f919bfe8573","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale","work_id":"e96730e3-129b-4db6-b981-15ab7932e297","ref_index":3,"cited_arxiv_id":"2010.11929","is_internal_anchor":true},{"doi":"","year":2021,"title":"An image is worth 16x16 words: Transformers for image recognition at scale","work_id":"b2cc82a1-974a-4867-8a88-379d15de9985","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Customized segment anything model for medical image segmentation, 2023","work_id":"472dad9d-9f7b-4c0e-8d78-095fc2ed9ae2","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":51,"snapshot_sha256":"5833864fd9efc22beeea4684c6c55cf8d7be8ab812c0ccdc7f4d2be8c8174791","internal_anchors":5},"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"}