Tanaka Kei, Lami Kris, Odate Takuma, Hori Takashi, Sato Tsubasa, Okoshi Ethan N, Fukuoka Junya
Department of Pathology Informatics, Nagasaki University Graduate School of Biomedical Sciences, Nagasaki, Japan.
Department of Pathology, Kameda Medical Center, Chiba, Japan.
Diagn Cytopathol. 2025 Nov;53(11):568-575. doi: 10.1002/dc.70015. Epub 2025 Aug 29.
For applications of digital pathology in cytology, challenges such as focal precision and data volume remain. The goals of this validation study are to compare diagnostic accuracy, screening time, annotation counts, and inter- and intra-observer agreement between digital slides using Z-stack scanning (z-WSI) and conventional glass slides in liquid-based cervical cytology (LBC).
We collected 91 LBC samples, with an equal number of NILM, LSIL, HSIL, and SCC cases. Four cytotechnologists evaluated cases using glass slides and z-WSI separately. They classified cases under two separate schemas: (1) "Screening-2-Category": NILM (normal) vs. other lesions (ASC-US and above); and (2) "Morpho-3-Category": NILM vs. LSIL (mild dysplasia) vs. ASC-H and higher (moderate dysplasia to squamous cell carcinoma) to reflect lesion severity and treatment implications.
For Screening-2-Category classifications, inter-observer agreement was 0.685 for glass slides and 0.637 for z-WSI, with intra-observer agreement ranging from 82.4% to 95.6%. For Morpho-3-Category classifications, inter-observer agreement was 0.700 for glass slides and 0.598 for z-WSI, indicating reduced agreement with z-WSI. Accuracy was 91.2% (glass slides) and 87.1% (z-WSI) for Screening-2-Category, and 86.5% and 81.0% for Morpho-3-Category, with no significant differences. In both modalities, cytotechnologists tended to apply more annotations in true positive cases but fewer in false negative cases. Screening time for z-WSI was 2-5 min longer on average for all cytotechnologists.
z-WSI is not completely equivalent to glass slides, but it has the potential to be used as a tool for cytology screening. Training specifically designed for WSI is expected to enhance diagnostic accuracy and improve workflow efficiency.
对于数字病理学在细胞学中的应用，诸如局部精度和数据量等挑战依然存在。本验证研究的目的是比较在液基宫颈细胞学（LBC）中使用Z-stack扫描（z-WSI）的数字玻片与传统玻璃玻片之间的诊断准确性、筛查时间、注释数量以及观察者间和观察者内的一致性。
我们收集了91份LBC样本，非典型鳞状细胞意义不明确（NILM）、低度鳞状上皮内病变（LSIL）、高度鳞状上皮内病变（HSIL）和鳞状细胞癌（SCC）病例数量相等。四名细胞技术人员分别使用玻璃玻片和z-WSI评估病例。他们在两种不同的模式下对病例进行分类：（1）“筛查-2-分类”：NILM（正常）与其他病变（非典型鳞状细胞不除外高级别鳞状上皮内病变及以上）；（2）“形态学-3-分类”：NILM与LSIL（轻度发育异常）与非典型鳞状细胞不除外高级别鳞状上皮内病变及更高级别（中度发育异常至鳞状细胞癌），以反映病变严重程度和治疗意义。
对于“筛查-2-分类”，玻璃玻片的观察者间一致性为0.685，z-WSI为0.637，观察者内一致性范围为82.4%至95.6%。对于“形态学-3-分类”，玻璃玻片的观察者间一致性为0.700，z-WSI为0.598，表明z-WSI的一致性降低。“筛查-2-分类”的准确率玻璃玻片为91.2%，z-WSI为87.1%；“形态学-3-分类”的准确率分别为86.5%和81.0%，无显著差异。在两种模式下，细胞技术人员倾向于在真阳性病例中应用更多注释，而在假阴性病例中应用较少。所有细胞技术人员使用z-WSI的筛查时间平均长2 - 5分钟。
z-WSI与玻璃玻片并不完全等同，但它有潜力用作细胞学筛查工具。专门为WSI设计的培训有望提高诊断准确性并改善工作流程效率。
