[Performance evaluation and validation of automated digital image analysis in peripheral blood cells morphology examination].

Objective: To verify and evaluate the performance of automated digital image(DIA) for peripheral blood cell morphology examination. Methods: Three hundred and seventy-nine routine peripheral blood smears and 18 plasmodium positive peripheral blood smears were collected. Blood smears were made and stained by Wright -Giemsa method.White blood cell (WBC) differentiation of blood smears were pre-classified by DIA (DIA direct classification), re-classified (manually reviewed after DIA classification), and artificially classified under microscope. the inter-assay and intra-assay coefficients of variation (CV) of DIA were respectively calculated for repeatability verification. Taking the artificial microscopy as the gold standard, the sensitivity、specificity and accuracy of DIA were calculated. The DIA ability of peripheral blood blast cell morphological count, platelet (PLT) morphological count and morphological examination of plasmodium were also verified. Results: Except for eosinophils and basophils, the inter-assay and intra-assay CV of WBC classification by DIA in normal samples were < 10%. The CV of WBC classification in abnormal samples increased with the decrease of cell percentage. The sensitivity, specificity and accuracy of DIA pre-classification were 90.5%, 99.2%, 98.2%. Through pre-classification and re-classification by DIA,the results of the blood smears which triggered blast cell alarm had a good correlation with manual classification(r=0.812, 0.983, both P<0.01). The PLT morphological count by DIA had high correlation with hematology analyzer (r=0.946, P<0.01). The deviation absolute value of two methods of PLT count was < 15%, while in PLT aggregation or giant thrombocytosis samples,the deviation absolute value of PLT count by two methods was > 15%. After image acquisition by DIA, 17 plasmodium trophozoites were detected in 18 plasmodium-positive peripheral blood smears, and the images were clear. Conclusions: The DIA system has good repeatability, high sensitivity, specificity and accuracy in peripheral blood WBC classification. Its pre-classification and re-classification results have high correlation with the manual classification results.目的： 评价全自动数字图像分析（DIA）系统用于外周血细胞形态学分类计数的性能。 方法： 选取北京大学第一医院血常规检验外周血涂片379份，疟原虫感染外周血涂片18份，经瑞氏-姬姆萨染色，通过DIA系统完成白细胞（WBC）预分类（DIA直接分类）、再分类（DIA分类后人工审核），以及人工直接显微镜分类。分别计算DIA系统的批内及批间变异系数（CV）进行重复性验证；以人工分类结果为金标准，计算DIA系统检测的灵敏度、特异度、正确率。并对其外周血原始细胞形态学计数、血小板（PLT）形态学计数及疟原虫形态学检查能力进行验证。 结果： 除嗜酸粒细胞、嗜碱粒细胞外，DIA系统对正常样本WBC分类的CV均<10%，异常样本WBC分类的CV随所占百分比减低而增大；DIA系统进行WBC预分类的灵敏度、特异度和正确率分别为90.5%、99.2%和98.2%。触发原始细胞报警的血涂片经DIA系统预分类、再分类后，其结果均与人工分类结果具有相关性（r=0.812、0.983，均P<0.01）。DIA系统血小板形态学计数与血细胞分析仪计数结果具有相关性（r=0.946，P<0.01）；两种方法计数血小板偏差绝对值<15%；血小板聚集及巨大血小板增多样本偏差绝对值均>15%。经DIA系统采集图像的18份疟原虫阳性外周血涂片，其中17份检出疟原虫滋养体，且图像清晰。 结论： DIA系统进行外周血分类计数的重复性较好，灵敏度、特异度和正确率较高；其预分类、再分类结果与人工分类结果具有相关性。.