School of Electrical Engineering and Computer Science, The Pennsylvania State University, University Park, Pennsylvania 16802, United States.
Materials Research Institute, The Pennsylvania State University, University Park, Pennsylvania 16802, United States.
ACS Sens. 2020 Oct 23;5(10):3140-3149. doi: 10.1021/acssensors.0c01238. Epub 2020 Oct 6.
Rapid antibacterial susceptibility testing (RAST) methods are of significant importance in healthcare, as they can assist caregivers in timely administration of the correct treatments. Various RAST techniques have been reported for tracking bacterial phenotypes, including size, shape, motion, and redox state. However, they still require bulky and expensive instruments-which hinder their application in resource-limited environments-and/or utilize labeling reagents which can interfere with antibiotics and add to the total cost. Furthermore, the existing RAST methods do not address the potential gradual adaptation of bacteria to antibiotics, which can lead to a false diagnosis. In this work, we present a RAST approach by leveraging machine learning to analyze time-resolved dynamic laser speckle imaging (DLSI) results. DLSI captures the change in bacterial motion in response to antibiotic treatments. Our method accurately predicts the minimum inhibitory concentration (MIC) of ampicillin and gentamicin for a model strain of ( K-12) in 60 min, compared to 6 h using the currently FDA-approved phenotype-based RAST technique. In addition to ampicillin (a β-lactam) and gentamicin (an aminoglycoside), we studied the effect of ceftriaxone (a third-generation cephalosporin) on K-12. The machine learning algorithm was trained and validated using the overnight results of a gold standard antibacterial susceptibility testing method enabling prediction of MIC with a similarly high accuracy yet substantially faster.
快速抗菌药敏测试（RAST）方法在医疗保健中具有重要意义，因为它们可以帮助护理人员及时给予正确的治疗。已经报道了各种 RAST 技术来跟踪细菌表型，包括大小、形状、运动和氧化还原状态。然而，它们仍然需要庞大而昂贵的仪器——这阻碍了它们在资源有限的环境中的应用——或者使用可能干扰抗生素并增加总成本的标记试剂。此外，现有的 RAST 方法并不能解决细菌对抗生素逐渐适应的潜在问题，这可能导致误诊。在这项工作中，我们提出了一种基于机器学习的 RAST 方法，通过分析时分辨动态激光散斑成像（DLSI）结果来实现。DLSI 捕捉了细菌运动对抗生素治疗的反应变化。与目前 FDA 批准的基于表型的 RAST 技术相比，我们的方法可以在 60 分钟内准确预测氨苄西林和庆大霉素对模型菌株 (K-12) 的最小抑菌浓度（MIC），而使用该方法则需要 6 小时。除了氨苄西林（一种β-内酰胺）和庆大霉素（一种氨基糖苷类）外，我们还研究了头孢曲松（一种第三代头孢菌素）对 K-12 的影响。该机器学习算法使用金标准抗菌药敏测试方法的过夜结果进行训练和验证，从而能够以类似的高精度但速度更快地预测 MIC。
