文献速递泌尿与男性生殖病理细胞病理与分子病理

人工智能整合尿液细胞学提升尿路上皮癌诊断:活检验证的准确性与效率增益

Enhancing urothelial carcinoma diagnosis with artificial intelligence-integrated urine cytology: Biopsy-validated accuracy and efficiency gain.
PubMed 文献 · Cancer Cytopathol · 2026-01-01

该回顾性研究评估AI辅助数字尿液细胞学系统AIxURO在高通量美国医疗中心的诊断性能与工作流程影响。200例ThinPrep玻片经两种定制扫描仪数字化,三名复核者独立评估三种模式。AIxURO较显微镜提高敏感性(85.0%和88.3% vs. 79.3%)和阴性预测值(85.1%和87.6% vs. 82.0%),但特异性(85.7%和82.3% vs. 94.3%)和阳性预测值(85.6%和83.3% vs. 93.3%)降低,中位复核时间缩短66%-78%(p < .0001)。

要点

  • AIxURO提高尿路上皮癌检测敏感性和阴性预测值,但降低特异性和阳性预测值。
  • AIxURO将中位复核时间缩短66%-78%(p < .0001),提升筛查效率。
  • 在98例活检确认病例和16例血尿活检病例中,AI辅助筛查显示更优诊断性能。

核心结论

  • AIxURO增强尿路上皮癌检测敏感性并显著提高筛查效率,但相较显微镜降低特异性和阳性预测值。

研究设计与样本

  • 回顾性研究,200例ThinPrep尿液细胞学玻片经两种定制扫描仪(AIS-1和AIS-2)数字化,三名复核者独立评估显微镜、AIxURO-AIS-1和AIxURO-AIS-2三种模式。

主要发现

  • AIxURO(AIS-1和AIS-2)敏感性为85.0%和88.3%,阴性预测值为85.1%和87.6%,显微镜分别为79.3%和82.0%。
  • AIxURO特异性为85.7%和82.3%,阳性预测值为85.6%和83.3%,显微镜分别为94.3%和93.3%。
  • AIxURO将中位复核时间缩短66%-78%(p < .0001)。
  • 200例中98例(49%)有活检确认,AIxURO显示较显微镜更高的敏感性和阴性预测值。
  • 16例血尿活检病例中,AI辅助筛查达到更优诊断性能。

临床怎么用

  • AIxURO可用于提高尿液细胞学筛查敏感性和效率,但需注意特异性降低可能增加假阳性。

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: Urine cytology is a noninvasive tool for detecting urothelial carcinoma, yet its performance depends heavily on expert cytologists and time-intensive glass slide review. Artificial intelligence (AI)-assisted digital cytology has emerged as a potential solution to improve diagnostic sensitivity and workflow efficiency. This retrospective study aimed to evaluate the diagnostic performance and workflow impact of AIxURO, an AI-assisted digital urine cytology system, in a high-volume US medical center. METHODS: Two hundred ThinPrep cytology slides were digitized with two distinct customized scanners integrated into the AIxURO imaging system (AIS-1 and AIS-2). Three reviewers independently assessed each case across three diagnostic modalities: microscopy, AIxURO-AIS-1, and AIxURO-AIS-2. Performance for binary bladder cancer detection was compared against expert consensus cytology diagnoses; review time, biopsy correlation, and performance in patients presenting with hematuria were also evaluated. RESULTS: AIxURO (AIS-1 and AIS-2) improved diagnostic sensitivity (85.0% and 88.3%) and negative predictive value (NPV) (85.1% and 87.6%) relative to microscopy (79.3% and 82.0%), whereas accuracy remained comparable across modalities. These sensitivity gains were accompanied by lower specificity (85.7% and 82.3% vs. 94.3%) and positive predictive value (PPV) (85.6% and 83.3% vs. 93.3%) compared with microscopy. Both AI modalities reduced classifying atypical urothelial cells and above cases as negative for high-grade urothelial carcinoma cases, and shortened median review time by 66%-78% (p < .0001). Among the 200 cases, 98 (49%) had biopsy confirmation, in which AIxURO showed higher sensitivity and NPV than microscopy. In 16 hematuria cases with biopsy correlation, AI-assisted screening achieved superior diagnostic performance. CONCLUSIONS: AIxURO enhances sensitivity for detecting urothelial carcinoma and markedly improves screening efficiency, with a tradeoff of reduced specificity/PPV compared with microscopy.

来源:Cancer Cytopathol,查看原文。版权归原作者所有。

引用

  1. Lajara S, Cuda J, Geisler DL, et al. Enhancing urothelial carcinoma diagnosis with artificial intelligence-integrated urine cytology: Biopsy-validated accuracy and efficiency gain. Cancer Cytopathol. 2026. PMID: 42144873. doi:10.1002/cncy.70110

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。