Abstract: Objective　To evaluate the effectiveness of an artificial intelligence (AI)⁃enabled microscopic imaging recognition system integrated in the modified Kato⁃Katz thick smear technique for detection of Schistosoma japonicum eggs, so as to provide insights into precise control and elimination of schistosomiasis. Methods　In October 2023, 20 fecal samples were collected from healthy residents negative for S. japonicum infection in Wuhan City, and each fecal sample was prepared into 4 Kato⁃Katz test slides, with 3 slides added S. japonicum egg suspensions with concentrations of approximately 25, 10, and 5 eggs per 10 μL, respectively, and one untreated. A total of 80 Kato⁃Katz test slides were prepared, and were divided into mild, moderate, and severe infection groups, and a negative control group, according to the number of eggs per gram of feces on each slide, with 20 slides in each group. S. japonicum eggs were detected on 80 Kato⁃Katz test slides with the AI⁃enabled microscopic imaging recognition system and manual microscopy, and the differences were compared between the two methods in terms of average detection time, accurate rate of qualitative detection, accurate rate of quantitative detection, percentage of missed detection, and percentage of false detection. Results　The average detection time of the imaging recognition system was longer than manual microscopy for detection of S. japonicum eggs on Kato⁃Katz test slides in all groups [(16.70 ± 0.01) min vs. (15.78 ± 2.11) min; t = 3.90, P < 0.05]. The detection time of the imaging recognition system was shorter than manual microscopy for detection of S. japonicum eggs on Kato⁃Katz test slides in the severe infection group (t = -3.91, P < 0.05), but was longer than manual microscopy in the the mild infection group (t = 5.03, P < 0.05) and the negative control group (t = 8.37, P < 0.05), while there was no significant difference in the detection time between the two methods in the moderate infection group (t = -0.09, P > 0.05). In addition, the imaging recognition system [97.50% (78/80) and 91.67% (55/60)] had higher accurate rates of both qualitative and quantitative detections than manual microscopy [81.25% (65/80) and 31.67% (19/60)] ([χ2] = 11.08 and 34.11, both P values < 0.05), and the imaging recognition system had a lower percentage of missed detection in the infection groups [3.33% (2/60)] and a lower percentage of false detection in the negative control group (0) than manual microscopy [13.33% (8/60) and 35.00% (7/20)] ([χ2] = 6.07, 5.14, both P values < 0.05). Conclusions　The AI⁃enabled microscopic imaging recognition system is effective to improve the accuracy for detection of S. japonicum eggs with the Kato⁃Katz technique, and is accurate to quantify and simple to perform, which may provide technical support for diagnosis of schistosomiasis and other parasitic diseases.

Key words: Schistosoma japonicum, Egg, Artificial intelligence, Microscopic imaging recognition system, Modified Kato?Katz thick smear technique, Etiological detection

摘要： 目的　评价在改良加藤厚涂片（Kato⁃Katz）法中使用基于人工智能（artificial intelligence，AI）的显微成像识别系统检测血吸虫虫卵的效果，为血吸虫病精准防控和消除提供新思路。方法　2023年10月，采集20名武汉市健康居民血吸虫感染阴性粪便样本共20份，取每份粪便样本制作4张Kato⁃Katz测试片，其中3张分别加入虫卵浓度约为25、10个/10 μL和5个/10 μL的血吸虫虫卵悬液，1张不做处理。共制备80张Kato⁃Katz测试片，根据各测试片的每克粪便虫卵数，将其分为轻度、中度、重度感染组和阴性对照组，每组20张。分别采用基于AI的显微成像识别系统法（成像法）和人工镜检法（人工法）对80张Kato⁃Katz测试片进行检测，比较两种方法平均检测时间、定性正确率、定量准确率、漏检率和误检率的差异。结果　成像法检测各组Kato⁃Katz测试片的平均时间［（16.70 ± 0.01）min］长于人工法［（15.78 ± 2.11）min］，差异有统计学意义（t = 3.90，P < 0.05）。进一步分析发现，成像法对重度感染组测试片的检测时间短于人工法（t = -3.91，P < 0.05），对轻度感染组（t = 5.03，P < 0.05）和阴性对照组（t = 8.37，P < 0.05）的检测时间长于人工法，差异均有统计学意义；但在中度感染组中，两种方法检测时间差异无统计学意义（t = -0.09，P > 0.05）。成像法的定性正确率［97.50%（78/80）］和定量准确率［91.67%（55/60）］均高于人工法［81.25%（65/80）、31.67%（19/60）］，差异均有统计学意义（[χ2]  = 11.08、34.11，P均< 0.05）；成像法对阳性Kato⁃Katz测试片的漏检率［3.33%（2/60）］和对阴性Kato⁃Katz测试片的误检率（0）均低于人工法［13.33%（8/60）、35.00%（7/20）］，差异均有统计学意义（[χ2] = 6.07、5.14，P均< 0.05）。结论　基于AI的显微成像识别系统不仅操作简单，而且还能提高Kato⁃Katz法检测血吸虫虫卵的正确率，并能实现准确定量。该方法有望为血吸虫病及其他寄生虫病诊断提供技术支持。

关键词: 血吸虫, 虫卵, 人工智能, 显微成像识别系统, 改良加藤厚涂片法, 病原学检测