{"paper":{"title":"Keyed Nonlinear Transform: Lightweight Privacy-Enhancing Feature Sharing for Medical Image Analysis","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A keyed nonlinear transform applied to split-inference features cuts re-identification AUC from 0.635 to 0.586 with 0.15 ms overhead and no backbone retraining.","cross_cats":["cs.CV"],"primary_cat":"eess.IV","authors_text":"Gyeongjung Kim, Haebom Lee","submitted_at":"2026-05-13T21:17:41Z","abstract_excerpt":"Feature sharing via split inference offers a lightweight alternative to federated learning for resource-constrained hospitals, but transmitted features still leak patient identity information and lack practical mechanisms for controlled feature sharing. We propose Keyed Nonlinear Transform (KNT), a drop-in feature transformation that applies key-conditioned obfuscation to intermediate representations. KNT reduces re-identification AUC from 0.635 to 0.586, corresponding to a 36% reduction in above-chance identity signal, while introducing only 0.15 ms CPU overhead, without backbone retraining, "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"KNT reduces re-identification AUC from 0.635 to 0.586, corresponding to a 36% reduction in above-chance identity signal, while introducing only 0.15 ms CPU overhead, without backbone retraining, and preserving classification performance within 1.0 pp.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that the secret key stays secure and that the nonlinear transform forces any inversion attempt into iterative gradient-based optimization even under full key compromise.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"KNT applies key-conditioned nonlinear obfuscation to split-inference features, cutting re-identification AUC from 0.635 to 0.586 with 0.15 ms overhead and under 1 pp accuracy loss.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A keyed nonlinear transform applied to split-inference features cuts re-identification AUC from 0.635 to 0.586 with 0.15 ms overhead and no backbone retraining.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"37f09880d8a8ca9d4690771777d08529315f0be32650ae21d106974ddd5f40f7"},"source":{"id":"2605.14123","kind":"arxiv","version":1},"verdict":{"id":"ec11c5a2-1803-4d53-b2e2-02cb1701ac70","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T02:09:01.475946Z","strongest_claim":"KNT reduces re-identification AUC from 0.635 to 0.586, corresponding to a 36% reduction in above-chance identity signal, while introducing only 0.15 ms CPU overhead, without backbone retraining, and preserving classification performance within 1.0 pp.","one_line_summary":"KNT applies key-conditioned nonlinear obfuscation to split-inference features, cutting re-identification AUC from 0.635 to 0.586 with 0.15 ms overhead and under 1 pp accuracy loss.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that the secret key stays secure and that the nonlinear transform forces any inversion attempt into iterative gradient-based optimization even under full key compromise.","pith_extraction_headline":"A keyed nonlinear transform applied to split-inference features cuts re-identification AUC from 0.635 to 0.586 with 0.15 ms overhead and no backbone retraining."},"references":{"count":27,"sample":[{"doi":"10.1145/2976749.2978318","year":2016,"title":"Brendan and Mironov, Ilya and Talwar, Kunal and Zhang, Li , year=","work_id":"518424d2-1085-4f06-9f9f-1c3aa7913ecb","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2026,"title":"Privacy-preserving collaborative medical image segmentation using la- tent transform networks.arXiv preprint arXiv:2603.05541, 2026","work_id":"964979c6-9e29-43c7-8566-98a9a3df4d6d","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Is pri- vate learning possible with instance encoding? InProceedings of the 2021 IEEE Sym- posium on Security and Privacy (S&P), pages 410–427","work_id":"81b6d1b5-a55f-4b5d-ba66-48f45efe4021","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Skin lesion analysis toward melanoma detection 2018: A challenge hosted by the international skin imaging collaboration (isic)","work_id":"a1680554-54cc-449f-aa92-0acb940cad0a","ref_index":4,"cited_arxiv_id":"1902.03368","is_internal_anchor":true},{"doi":"10.1561/0400000042","year":2014,"title":"URLhttps://doi.org/10.1561/ 0400000042","work_id":"2b98a090-16f8-424b-b782-6b54235093c7","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":27,"snapshot_sha256":"2603611d3c9f75c232666e61c890c5275f935c6e18cea3e6c5a7d398d7bad2ea","internal_anchors":2},"formal_canon":{"evidence_count":2,"snapshot_sha256":"d8bb4c1e4096b7d3f13c9a16d4a1f62abad54e738bf88eaf64efb4bc742c6d10"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}