{"paper":{"title":"Warm-Started Reinforcement Learning for Iterative 3D/2D Liver Registration","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A warm-started RL policy performs iterative 6-DoF CT-to-video liver registration and learns its own stopping criterion, reaching 15.70 mm TRE without extra optimization.","cross_cats":["physics.med-ph"],"primary_cat":"cs.CV","authors_text":"Abdolrahim Kadkhodamohammadi, Brian R. Davidson, Danail Stoyanov, Evangelos B. Mazomenos, Hanyuan Zhang, Lucas He, Matthew.J Clarkson, Zijie Cheng","submitted_at":"2026-04-11T14:58:45Z","abstract_excerpt":"Registration between preoperative CT and intraoperative laparoscopic video plays a crucial role in augmented reality (AR) guidance for minimally invasive surgery. Learning-based methods have recently achieved registration errors comparable to optimization-based approaches while offering faster inference. However, many supervised methods produce coarse alignments that rely on additional optimization-based refinement, thereby increasing inference time.\n  We present a discrete-action reinforcement learning (RL) framework that formulates CT-to-video registration as a sequential decision-making pro"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experiments on a public laparoscopic dataset demonstrated that our method achieved an average target registration error (TRE) of 15.70 mm, comparable to supervised approaches with optimization, while achieving faster convergence.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The warm-started RL policy can reliably learn effective 6-DoF rigid transformations and a stopping criterion from the shared encoder features without post-hoc tuning or overfitting to the specific dataset.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A warm-started discrete-action RL framework for CT-to-video liver registration achieves 15.70 mm average TRE with faster convergence than supervised methods plus optimization.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A warm-started RL policy performs iterative 6-DoF CT-to-video liver registration and learns its own stopping criterion, reaching 15.70 mm TRE without extra optimization.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"771a98c4eccbca92b4dbc402172d845fef535e5af946b0c296ca0731b934d31e"},"source":{"id":"2604.10245","kind":"arxiv","version":2},"verdict":{"id":"e9b7ba05-1384-4ca4-9e96-69e158994c22","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T16:28:53.192594Z","strongest_claim":"Experiments on a public laparoscopic dataset demonstrated that our method achieved an average target registration error (TRE) of 15.70 mm, comparable to supervised approaches with optimization, while achieving faster convergence.","one_line_summary":"A warm-started discrete-action RL framework for CT-to-video liver registration achieves 15.70 mm average TRE with faster convergence than supervised methods plus optimization.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The warm-started RL policy can reliably learn effective 6-DoF rigid transformations and a stopping criterion from the shared encoder features without post-hoc tuning or overfitting to the specific dataset.","pith_extraction_headline":"A warm-started RL policy performs iterative 6-DoF CT-to-video liver registration and learns its own stopping criterion, reaching 15.70 mm TRE without extra optimization."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.10245/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"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"}