中文利用纵向上下文的临床医生验证交互式病灶追踪
ENExploiting Longitudinal Context in Clinician-Verified Interactive Lesion Tracking
arXiv:2605.23118v1 提出“验证跟踪”范式:临床医生验证配准提示后,模型利用该提示进行肿瘤病变跟踪,兼顾端到端跟踪的高自动化和配准-分割管线的可验证性,解决了自动跟踪失败不可纠正及忽略既往外观的局限,提升模糊病例的跟踪准确性。
arXiv:2605.23118v1 Announce Type: new Abstract: Tracking tumor lesions across serial CT scans is essential for oncological response assessment. Existing automated methods face a fundamental trade-off: end-to-end trackers achieve high automation but offer no opportunity to correct silent tracking failures, while decoupled registration-segmentation pipelines permit user verification yet discard the lesion's prior appearance, limiting accuracy in ambiguous cases. In this work, we propose a Verified Tracking paradigm: a clinician verifies a registration-proposed prompt, which the model leverages a