Tyndall National Institute, University College Cork, Lee Maltings Complex, Dyke Parade, Cork, T12R5CP, Ireland.
Department of Physiological Sciences, Federal University of Espírito Santo (UFES), Vitória, Brazil.
Sci Rep. 2021 Oct 11;11(1):15409. doi: 10.1038/s41598-021-93511-2.
Early diagnosis of COVID-19 in suspected patients is essential for contagion control and damage reduction strategies. We investigated the applicability of attenuated total reflection (ATR) Fourier transform infrared (FTIR) spectroscopy associated with machine learning in oropharyngeal swab suspension fluid to predict COVID-19 positive samples. The study included samples of 243 patients from two Brazilian States. Samples were transported by using different viral transport mediums (liquid 1 or 2). Clinical COVID-19 diagnosis was performed by the RT-PCR. We built a classification model based on partial least squares (PLS) associated with cosine k-nearest neighbours (KNN). Our analysis led to 84% and 87% sensitivity, 66% and 64% specificity, and 76.9% and 78.4% accuracy for samples of liquids 1 and 2, respectively. Based on this proof-of-concept study, we believe this method could offer a simple, label-free, cost-effective solution for high-throughput screening of suspect patients for COVID-19 in health care centres and emergency departments.
对疑似 COVID-19 患者进行早期诊断对于控制传染和减少损害策略至关重要。我们研究了衰减全反射（ATR）傅里叶变换红外（FTIR）光谱结合机器学习在口咽拭子悬浮液中对预测 COVID-19 阳性样本的适用性。该研究包括来自巴西两个州的 243 名患者的样本。样本通过使用不同的病毒运输介质（液体 1 或 2）进行运输。临床 COVID-19 诊断通过 RT-PCR 进行。我们建立了一个基于偏最小二乘（PLS）和余弦 K-最近邻（KNN）的分类模型。我们的分析结果分别为液体 1 和 2 的样本的 84%和 87%的敏感性、66%和 64%的特异性以及 76.9%和 78.4%的准确性。基于这项概念验证研究，我们认为该方法可以为医疗中心和急诊部门对疑似 COVID-19 患者进行高通量筛查提供一种简单、无标记、经济有效的解决方案。
