基于深度学习的人工智能系统在恶性原虫检测中的应用

Application of artificial intelligence system based on deep learning in the detection of Plasmodium falciparum

目的研究基于深度学习的人工智能(AI)技术在恶性症原虫检测中的应用价值。方法选取中国籍入境人员血涂片100张,其中恶性疟原虫病例阳性血涂片50张,阴性血涂片50张。将其打乱顺序后进入高清扫描仪后,数据自动导入AI工作站。采取AI阅片、人工独立阅片以及AI辅助阅片的方式,对恶性疮原虫进行检测。再将判读结果与恶性疟“金标准”对比,最后比较3种方法的正确度、灵敏度、特异度以及检测时间。结果AI辅助阅片正确度最高(0.95),AI次之(0.92),人工阅片最差(0.85)。检测时间AI用时最少(2.3min),AI辅助阅片次之(5.6min),人工阅片用时最多(12.8min)。结论结合阅片准确性(正确度、灵敏度以及特异度在90%以上)及用时(<10min)综合考虑,AI辅助人工阅片是实验室最佳方式,可提高恶性疟原虫检测的灵敏度及工作效率,基于深度学习的AI系统在恶性原虫人工智能检测中具有较高的应用价值,可在海关口岸实验室广泛应用。

Objective This paper aims to study the application value of artificial intelligence(AI)technology based on deep learning in the detection of Plasmodium falciparum.Methods A total of 100 blood smears were selected from inbound Chinese nationals,including 50 positive blood smears of Plasmodium falciparum cases and 50 negative blood smears.After entering the HD scanner randomly,the data were automatically imported into the AI workstation.AI image reading,artificial image reading and AI assisted image reading were used to detect Plasmodium falciparum.The interpretation results were compared with the'gold standard"of falciparum malaria,and the accuracy,sensitivity,specificity and detection time of the three methods were compared.Results The accuracy of AI assisted image reading was the highest(0.95),followed by AI image reading(0.92)and artificial image reading(0.85).The detection time of AI image reading was the least(2.3 min),followed by AI assisted image reading(5.6 min)and artificial image reading(12.8 min).Conclusion Considering the reading efficiency(accuracy,sensitivity and specificity>90%)and time(<10 min),AI assisted image reading is the best method in the laboratory,which can improve the sensitivity and efficiency of Plasmodium falciparum detection.AI system based on deep learning has high application value in the artificial intlligence detection of Plasmodium falciparum,which can be widely used in customs laboratories.