医学
骨髓纤维化
核医学
病理
随机对照试验
外周血
放射科
技术员
形态学(生物学)
再现性
流式细胞术
血膜
平均差
多中心研究
作者
Song‐Song Lu,Yonghui Guo,Dong-Mei Liu,Xiao-ning CHEN,Meng-jie ZHU,Zhong-hui ZHANG,Wei Zhai,Man HAN,Zhao Jing-zhong,Yi Ye,Jianping Xiao,Yu-jing CHEN,Ya-ying LIN,Jie-yi LU,Hui-Xian Luo,Jun Cao,Yong WANG,Dong Zheng,Xuedong Chen,Jun QIU
标识
DOI:10.1038/s41746-026-03089-0
摘要
Peripheral blood smear examination is labor-intensive and operator-dependent. We conducted a multicenter, prospective, randomized paired study of an AI-assisted digital morphology system in routine practice. Among 1570 analyzed smears from three centers, AI-assisted analysis outperformed manual microscopy across 15 nucleated cell categories, including improved screening for acute promyelocytic leukemia through detection of abnormal promyelocytes (F1 score 0.794 [95% CI, 0.667–0.892] vs 0.658 [0.508–0.767]). AI assistance also improved diagnostic performance across most erythrocyte morphology categories, including better detection of teardrop cells, a morphologic clue relevant to myelofibrosis (F1 score 0.306 [0.233–0.374] vs 0.261 [0.190–0.323]). Platelet estimation was more accurate with AI assistance, with lower absolute mean deviation from flow cytometry (2.05 vs 14.08 × 10 9 /L). AI assistance increased technician efficiency by approximately 60%, supporting its potential role in improving the consistency and efficiency of morphology reporting for the evaluated cell categories in routine smear-review workflows.
科研通智能强力驱动
Strongly Powered by AbleSci AI