Bound-Aware Per-Organ Recall Risk Control for Multi-Organ CT Segmentation under Clinical Domain Shift
Bound-Aware Per-Organ Recall Risk Control for Multi-Organ CT Segmentation under Clinical Domain Shift
面向临床域偏移的多器官CT分割:基于边界感知的器官级召回风险控制
Abstract: Distribution-free risk control adds organ-specific recall guarantees to frozen segmentation. We calibrate per-organ thresholds for an AMOS-trained nnU-Net, audit transfer to RAOS, and estimate local re-certification cost using case-level voxel false-negative rate (FNR). The AMOS control passes, but $7/12$ organs exceed $\alpha{=}0.10$ after transfer; smaller calibration sets can mask exceedances with conservative or vacuous thresholds.
摘要: 无分布风险控制(Distribution-free risk control)为冻结的分割模型增加了器官级的召回率保证。我们为在 AMOS 数据集上训练的 nnU-Net 校准了各器官的阈值,审计了其向 RAOS 数据集的迁移情况,并利用病例级的体素假阴性率(FNR)估算了局部重新认证的成本。结果显示,AMOS 上的控制表现良好,但在迁移后,12 个器官中有 7 个的风险超过了 $\alpha{=}0.10$;较小的校准集可能会因保守或无效的阈值而掩盖这些超标现象。
Risk-Controlling Prediction Sets (RCPS) give high-probability control of population-mean risk, whereas Conformal Risk Control (CRC) gives weaker expectation control. Both require exchangeability; fixed and global thresholds give no per-organ guarantee. The Waudby—Smith—Ramdas (WSR) betting bound re-certifies six Tier-1 organs with 25 local cases, versus 30—40 for Hoeffding—Bentkus (HB). CRC needs 10—15 but has a heavier individual-case tail. No Tier-2 organ meets our illustrative precision criterion with 25 cases.
风险控制预测集(RCPS)提供了对总体平均风险的高概率控制,而一致性风险控制(CRC)提供的则是较弱的期望控制。两者都要求数据具有可交换性;而固定的全局阈值无法提供任何器官级的保证。Waudby—Smith—Ramdas (WSR) 投注界限法仅需 25 个局部病例即可重新认证 6 个一级器官,而 Hoeffding—Bentkus (HB) 方法则需要 30—40 个病例。CRC 虽然仅需 10—15 个病例,但在个体病例的尾部风险上表现较重。在 25 个病例的情况下,没有任何二级器官能达到我们设定的演示精度标准。
Paper Details:
- Authors: Souraj Adhikary, Negar Chabi, Andre Mastmeyer
- Submission Date: 18 Aug 2026
- Primary Category: Computer Vision and Pattern Recognition (cs.CV)
- DOI: 10.48550/arXiv.2608.18193
论文详情:
- 作者: Souraj Adhikary, Negar Chabi, Andre Mastmeyer
- 提交日期: 2026年8月18日
- 主要分类: 计算机视觉与模式识别 (cs.CV)
- DOI: 10.48550/arXiv.2608.18193