Silawan Nawatt, Kusukame Koichi, Kek Khai Jun, Kuan Win Sen
Annu Int Conf IEEE Eng Med Biol Soc. 2018 Jul;2018:1612-1615. doi: 10.1109/EMBC.2018.8512541.
We propose a novel concept for core body temperature estimation to improve sensitivity and specificity of a fever screening system under different environmental conditions based on an infrared thermal camera. The conventional approach of setting low temperature thresholds to determine presence of fever to increase sensitivity has led to highly-degraded specificity due to the low accuracy in core body temperature estimation. Two main causes are the moderate correlation between core body temperature and surface temperature data used to determine it, and the estimation algorithm that does not consider changes in the environment. Hence, in our novel concept, we eliminate the environmental effects by using direct and correcting temperature data, and thus improve the accuracy in estimating core body temperature. The direct data contain rich information about core body temperature through maximum temperatures obtained from the mouth, ear, around the eye and forehead, while the correcting data contain information related to the surroundings such as the cheek and nose temperatures to compensate for the environmental effect on the former. Since direct data can be easily affected by the environment and noise, multiple direct data are taken to minimize this problem. Through improved accuracy, both sensitivity and specificity will be automatically increased and the trade-off between them when adjusting the threshold values will be greatly relaxed. Analysis of the results shows improvement in both sensitivity and specificity from78.9% and 87.0%, respectively, in the conventional approach, to 84.2% and 91.3% in the proposed method when 37.5$^{o}\text{C}$ was set as the threshold. Data in the present study was obtained from a wide spectrum of ages (between 22 and 58 years), ethnicities (seven) and core body temperatures (36.0$^{o}\text{C}$ to 39.5$^{o}\text{C}$). Data were also collected at variable room temperatures ranging from 20.2$^{o}\text{C}$ to 30.8$^{o}\text{C}$.
我们提出了一种用于核心体温估计的新颖概念，以提高基于红外热像仪的发热筛查系统在不同环境条件下的灵敏度和特异性。传统方法通过设置较低的温度阈值来确定是否发热以提高灵敏度，但由于核心体温估计的准确性较低，导致特异性大幅下降。两个主要原因是用于确定核心体温的核心体温与表面温度数据之间的相关性一般，以及估计算法未考虑环境变化。因此，在我们的新颖概念中，我们通过使用直接温度数据和校正温度数据来消除环境影响，从而提高核心体温估计的准确性。直接数据通过从口腔、耳朵、眼睛周围和额头获得的最高温度包含有关核心体温的丰富信息，而校正数据包含与周围环境相关的信息，如脸颊和鼻子温度，以补偿环境对前者的影响。由于直接数据很容易受到环境和噪声的影响，因此采用多个直接数据来最小化这个问题。通过提高准确性，灵敏度和特异性将自动提高，并且在调整阈值时它们之间的权衡将大大缓解。结果分析表明，当将37.5$^{o}\text{C}$设置为阈值时，灵敏度和特异性分别从传统方法中的78.9%和87.0%提高到了所提出方法中的84.2%和91.3%。本研究中的数据来自广泛的年龄范围（22至58岁）、种族（七个）和核心体温（36.0$^{o}\text{C}$至39.5$^{o}\text{C}$）。数据也是在20.2$^{o}\text{C}$至30.8$^{o}\text{C}$的可变室温下收集的。
