Cancer cytopathology · 2026
Artificial intelligence-assisted urine cytology based on the Paris System for Reporting Urothelial Carcinoma.
СтатьяПоддержка исследования: вне правительства СШАНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Chen Lan, Cao Xinyi, Lu Zongyue, Liu Longteng, Sun Mingjun, Wu Yulong, Zhang Wei
Аннотация
BACKGROUND: Urine cytology is a noninvasive and valuable tool for detecting urothelial carcinoma but suffers from variable sensitivity and observer dependency. Artificial intelligence (AI) may enhance the diagnostic accuracy and efficiency of urine cytology. The objective of this study was to develop and validate an AI-based cytology system for urothelial carcinoma detection in both clinical and screening contexts.
METHODS: In total, 328 retrospective clinical cases and 1489 prospective health screening samples were analyzed. All specimens underwent liquid-based cytology and were digitized at ×20 magnification. For model development, ed 269 annotated training slides (56,710 cells) were used. The AI pipeline mimicked cytopathologist workflow, integrating deep learning-based cell detection and segmentation with feature extraction and support vector machine classification into AI-negative, AI-atypical, and AI-positive categories according to The Paris System for Reporting the Urinary Cytology. Agreement was assessed using weighted κ values and prevalence-adjusted, bias-adjusted κ values.
RESULTS: In the clinical cohort, AI achieved 83.0% agreement with histopathology and 82.6% agreement with expert cytology. Weighted κ values (κ, 0.631-0.661) reflected substantial agreement, with discrepancies mainly between adjacent categories. In the health screening cohort, the prevalence-adjusted, bias-adjusted κ was 0.647 despite low prevalence. AI demonstrated a high negative predictive value of 99.7% and negative percent agreement of 82.3%, with a minimal false-omission rate (0.25%).
CONCLUSIONS: AI-assisted urine cytology exhibited substantial concordance with expert interpretation and histopathologic standards. Its capability of screen-out and high negative predictive value support its potential as a reliable triage tool, improving workflow efficiency while maintaining diagnostic reliability in large-scale screening.
METHODS: In total, 328 retrospective clinical cases and 1489 prospective health screening samples were analyzed. All specimens underwent liquid-based cytology and were digitized at ×20 magnification. For model development, ed 269 annotated training slides (56,710 cells) were used. The AI pipeline mimicked cytopathologist workflow, integrating deep learning-based cell detection and segmentation with feature extraction and support vector machine classification into AI-negative, AI-atypical, and AI-positive categories according to The Paris System for Reporting the Urinary Cytology. Agreement was assessed using weighted κ values and prevalence-adjusted, bias-adjusted κ values.
RESULTS: In the clinical cohort, AI achieved 83.0% agreement with histopathology and 82.6% agreement with expert cytology. Weighted κ values (κ, 0.631-0.661) reflected substantial agreement, with discrepancies mainly between adjacent categories. In the health screening cohort, the prevalence-adjusted, bias-adjusted κ was 0.647 despite low prevalence. AI demonstrated a high negative predictive value of 99.7% and negative percent agreement of 82.3%, with a minimal false-omission rate (0.25%).
CONCLUSIONS: AI-assisted urine cytology exhibited substantial concordance with expert interpretation and histopathologic standards. Its capability of screen-out and high negative predictive value support its potential as a reliable triage tool, improving workflow efficiency while maintaining diagnostic reliability in large-scale screening.