[论文] GazeRIB: Systematic Evaluation of Multi-Reader Gaze-Guided Chest X-ray...

论文概要

研究领域: CV 作者: Nattakorn Kittisut, Roy M. Gabriel, Marly van Assen, Mohammadreza Zandehshahvar, Carlo N. De Cecco, Elizabeth A. Krupinski, Ali Adibi 发布时间: 2026-10-30 arXiv: 2610.42152

中文摘要

放射科医生的视觉注意力可以监督医学影像学习，但阅读小组的构成如何影响其价值仍未得到验证。我们评估了 GazeRIB（Gaze-guided Radiologist-Informed Blackout，凝视引导的放射科医师知情遮蔽），该方法在训练期间遮蔽未观看的区域，并在推理时使用未遮蔽的图像。八名专家反复阅读相同的80张胸部X光片（CXR）。我们在七种架构和10个随机种子下，比较了单个阅读者与三人、五人和八人小组，在独立的2,897张CXR队列上评估了910个模型用于四分类严重程度预测。每个小组都将宏观F1分数相比无凝视基线提高了5.33-7.47个百分点（pp；Bonferroni校正后p < 0.007）；而单个阅读者在该指标上均未达到显著性。三名阅读者即可捕获八人小组在宏观F1、ROC-AUC和AU-PRC上73-80%的收益。在无需重新训练的外部验证中，所有四个小组在SIIM-FISABIO-RSNA数据集上将ROC-AUC提高了+1.93至+4.24个百分点（p ≤ 0.004）；在MIMIC-Eye上，所有小组与基线无显著差异，而单阅读者平均下降（-1.40 pp，p = 0.008）。因此，阅读者构成影响了性能，尽管迁移效果取决于具体队列。

原文摘要

Radiologists' visual attention can supervise medical image learning, but how reader panel composition affects its value remains untested. We evaluated GazeRIB (Gaze-guided Radiologist-Informed Blackout), which masks unviewed regions during training and uses unmasked images at inference. Eight experts repeatedly read the same 80 chest X-rays (CXRs). We compared individual readers with three-, five- and eight-reader panels across seven architectures and 10 seeds, evaluating 910 models on a separate 2,897-CXR cohort for four-class severity prediction. Every panel improved macro F1 over the no-gaze baseline by 5.33-7.47 percentage points (pp; Bonferroni-adjusted p < 0.007); no individual reader reached significance on this endpoint. Three readers captured 73-80% of the eight-reader gain in mac...

#论文 #arXiv #CV #小凯