低倍视野快速目标检测:数字细胞学诊断准确性的可改变认知标志物
研究通过眼动追踪分析100名细胞技术员诊断30张数字细胞学图像,发现工作经验与诊断准确性无显著相关,低倍视野主对象总注视时间短是高准确性的唯一独立预测因子;28名学生经3个月培训后首次目标注视时间显著缩短,表明该认知标志物可通过标准教育获得。
要点
- 工作经验与数字细胞学诊断准确性无显著相关。
- 低倍视野主对象总注视时间短是高准确性的独立预测因子。
- 低倍视野检测效率可通过培训获得,是客观评估技能发展的指标。
核心结论
- 低倍视野目标检测效率比专业经验更能预测诊断准确性。
- 低倍视野效率是可改变的认知标志物,可通过标准教育获得。
- 量化眼动指标为评估技能发展和AI时代准备度提供客观手段。
研究设计与样本
- 第一阶段:100名细胞技术员(1-40年经验)诊断30张数字细胞学图像,使用眼动追踪分析兴趣区注视指标。
- 第二阶段:28名学生完成3个月细胞技术培训,比较培训前后指标。
主要发现
- 工作经验年限与诊断准确性无显著相关(r=0.189,p>.05)。
- 多变量分析显示低倍视野主对象总注视时间短是高准确性的唯一独立预测因子(p=.045)。
- 经验仅与对样本信息的注意力相关。
- 培训后学生首次目标注视时间大幅缩短(r=0.62-0.86),对正常背景的注意力下降(r=0.71)。
临床怎么用
- 在数字细胞学阅片中优先快速低倍视野目标检测而非 exhaustive 筛查。
- 可使用眼动追踪指标客观评估细胞技术员的技能发展。
局限性
- 第二阶段样本量较小(28名学生),需进一步验证。
- 研究为观察性设计,眼动指标与诊断准确性的因果关系未确定。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
BACKGROUND: Traditionally, cytology expertise has been equated with professional experience. However, the transition to whole-slide imaging and artificial intelligence (AI) necessitates a shift from exhaustive screening to rapid verification. The goal of this study was to identify cognitive biomarkers associated with diagnostic accuracy and evaluate their modifiability. METHODS: In phase 1, 100 cytotechnologists with 1-40 years of experience diagnosed 30 digital cytology images using eye-tracking. Gaze metrics across areas of interest were analyzed via nominal logistic regression. In phase 2, 28 students completed a 3-month cytotechnology training program. Pre- and post-training metrics were compared using Wilcoxon signed-rank tests and effect sizes (r). RESULTS: Years of experience showed no significant correlation with diagnostic accuracy (r = 0.189, p > .05). Multivariate analysis identified shorter total fixation duration on the "low-power field (LPF) main object" as the sole independent predictor of high accuracy (p = .045), suggesting a "pop-out" detection mechanism. Experience correlated only with attention to sample information. Post-training (phase 2), students' time to first target fixation decreased substantially (r = 0.62-0.86), whereas their attention to normal backgrounds decreased (r = 0.71). This demonstrates the rapid acquisition of expert-like selective attention. CONCLUSIONS: Efficiency in LPF target detection is a better predictor of diagnostic accuracy than professional experience. This study identifies LPF efficiency as a modifiable cognitive biomarker that can be acquired through standard education. Quantifying these gaze metrics provides an objective means of evaluating skill development and readiness for the AI era.
引用
- Abe N, Nishimura Y, Yamashita K, et al. Screening efficiency over experience: Rapid target detection in low-power field as a modifiable cognitive biomarker for diagnostic accuracy in digital cytology. Cancer Cytopathol. 2026. PMID: 42516004. doi:10.1002/cncy.70132
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。