深度学习卷积神经网络实现血培养革兰染色自动判读
该研究利用自动显微镜平台扫描未盖玻片的血培养革兰染色玻片,采集25,488张图像并生成100,213个图像块,训练卷积神经网络(CNN)进行分类。CNN在测试集上分类准确率达94.9%,各分类ROC曲线下面积均>0.98;在189张全玻片上新图像验证中,对革兰阳性链/成对球菌、革兰阳性簇状球菌和革兰阴性杆菌的敏感度分别为98.4%、93.2%、96.3%,特异度分别为75.0%、97.2%、98.1%。
要点
- CNN对血培养革兰染色图像块分类准确率达94.9%。
- 全玻片验证中革兰阴性杆菌敏感度96.3%、特异度98.1%。
- 算法可预筛并分类图像块以加速人工镜检。
核心结论
- 研究支持血培养革兰染色全自动分类方法的可行性验证。
- 算法擅长识别含微生物的图像块,可向技术人员呈现预筛分类图像块以加速涂片审阅。
研究设计与样本
- 使用自动显微镜平台以40×干物镜扫描未盖玻片玻片。
- 从常规临床工作中制备的阳性血培养革兰染色收集25,488张图像,生成100,213个图像块。
主要发现
- CNN模型在测试集图像块上分类准确率为94.9%。
- 各分类ROC曲线下面积均>0.98。
- 189张全玻片验证:革兰阳性链/成对球菌敏感度98.4%、特异度75.0%。
- 革兰阳性簇状球菌敏感度93.2%、特异度97.2%。
- 革兰阴性杆菌敏感度96.3%、特异度98.1%。
关键数值与推荐
- 分类类别包括革兰阳性簇状球菌、革兰阳性链/成对球菌、革兰阴性杆菌和背景(无细胞)。
- 这些类别被选为概念验证开发目标,因其与大多数血流感染相关。
临床怎么用
- 算法可预筛并分类图像块,供技术人员加速涂片审阅。
- 该概念可能扩展至临床实验室所有革兰染色判读活动。
局限性
- 研究为概念验证,需进一步临床验证。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
Microscopic interpretation of stained smears is one of the most operator-dependent and time-intensive activities in the clinical microbiology laboratory. Here, we investigated application of an automated image acquisition and convolutional neural network (CNN)-based approach for automated Gram stain classification. Using an automated microscopy platform, uncoverslipped slides were scanned with a 40× dry objective, generating images of sufficient resolution for interpretation. We collected 25,488 images from positive blood culture Gram stains prepared during routine clinical workup. These images were used to generate 100,213 crops containing Gram-positive cocci in clusters, Gram-positive cocci in chains/pairs, Gram-negative rods, or background (no cells). These categories were targeted for proof-of-concept development as they are associated with the majority of bloodstream infections. Our CNN model achieved a classification accuracy of 94.9% on a test set of image crops. Receiver operating characteristic (ROC) curve analysis indicated a robust ability to differentiate between categories with an area under the curve of >0.98 for each. After training and validation, we applied the classification algorithm to new images collected from 189 whole slides without human intervention. Sensitivity and specificity were 98.4% and 75.0% for Gram-positive cocci in chains and pairs, 93.2% and 97.2% for Gram-positive cocci in clusters, and 96.3% and 98.1% for Gram-negative rods. Taken together, our data support a proof of concept for a fully automated classification methodology for blood-culture Gram stains. Importantly, the algorithm was highly adept at identifying image crops with organisms and could be used to present prescreened, classified crops to technologists to accelerate smear review. This concept could potentially be extended to all Gram stain interpretive activities in the clinical laboratory.
引用
- Smith KP, Kang AD, Kirby JE, et al. Automated Interpretation of Blood Culture Gram Stains by Use of a Deep Convolutional Neural Network. J Clin Microbiol. 2018. PMID: 29187563. doi:10.1128/JCM.01521-17
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。