Accuracy Assessment of the Morphological Analysis System with Automated Microscopy and Artificial Intelligence for Gram-stained Vaginal Discharge Smears

  Objective  To assess the accuracy of an automated microscopy system for the morphological examination of Gram-stained vaginal discharge smears.

  Methods  The specimens of vaginal discharge were collected from outpatients of the clinic of Obstetrics and Gynecology in Peking Union Medical College Hospital from January to May 2020. All of the smears were Gram-stained, observed by both manual and automated microscopy. The automated microscopy was made by a Comet-60au high power microscope and the morphological analyzing system with artificial intelligence. The accuracy assessment of automated microscopy for analyzing the morphology of vaginal micro-ecology was evaluated by the semi-quantitative score, in which manually microscopic results were used as control.

  Results  In total, 400 patients were enrolled. By manual microscopy, 124 (31.0%), 142 (35.5%) and 134 (33.5%) patients were diagnosed as bacterial vaginosis (BV), intermediate BV, and normal by Nugent scores, respectively. Yeast-like spores or pseudohyphae were found in smears of 137 (34.3%) patients by manual microcopy, in which the percentage of BV, intermediated BV, and normal Nugent-scored patients were 29 (21.2%), 68 (49.6%), and 40 (29.2%), respectively. Based on the semi-quantitative score of microscopic examination results, the basic and the complete concordance rate of vaginal flora diversity between automated microscopy and manual microscopy were 52.3% and 85.3%, respectively, and those of vaginal flora intensity, lactobacillus morphotypes, Gardnerella and Bacteroides spp .morphotypes and curved gram-variable rods were 81.8%-99.5% and 93.3%-100%. Their qualitative consistency rates of yeast-like spores and pseudohyphae were both higher than 96%. The complete concordance rate of more than ten white cells per high power-field was 70.5%. Compared to manual microscopy, the consistency rates of automated microscopy for BV and fungal qualitative examination were 90.8% and 94.5%, 0.861 and 0.875 with Kappa values (P < 0.001), respectively.

  Conclusions  In the morphological evaluation of Gram-stained vaginal discharge smears, automated microscopy showed good concordance with manual microscopy, which can supply a rapid and accurate clinical test, highly improve the testing efficiency, and reduce the load of manual work.