Methods to evaluate sensitivity of biomedical thermographic systems for body temperature determination

Abstract

Thermographic fever-screening systems have been deployed widely for the non-contact detection of febrile temperatures during the COVID-19 pandemic. The standard for required performance (IEC 80601-2-59:2017) [1] of thermographic screening systems describe a specification-based requirement to assure the system is able to detect real febrile temperatures, but does not describe any clinical testing. Thermographic systems have been shown to be sensitive to febrile temperatures in controlled [2,3] and real-world conditions [4]. However, concerns have been raised about elements of these standards [5,6]. In a related report by the author [20], the size-of-source artifact was presented as a major confound in thermographic systems, causing surface temperatures to be altered by enough to make certain systems miss common fever thresholds, and this confound is not addressed by the standard. More concerningly is the recent observation of unacceptably strong bias-towards-normal algorithms in a selection of widely-deployed thermographic systems [7]. On the other hand, the standard covering non-contact body thermometry (IEC 80601-2-56:2017) [8] describes both laboratory and clinical testing to assure the system can detect febrile temperatures. Many single-pixel and at least one multi-pixel infrared thermometers are widely available with 510(k) marketing approval by the US FDA. In principle, clinical testing can give a greater certainty about real-world device performance. However, as we show in this report, many if not most of these devices also resort to unacceptably strong bias-to-normal algorithms, with parameters such that they would not be able to distinguish body temperatures ranging from 95F to 103F from normal, similar to the observations of [7]. A challenge for the clinical testing of thermometry and screening devices is the impracticality of finding a population of individuals having a sufficient distribution of elevated temperatures, rather than one group of normal and one group of severely febrile. In this report, we characterize the physiology of the core-to-surface skin temperature relationship, show how several approved devices deviate significantly from this relation and demonstrate a simple test protocol to assess the real-world sensitivity and specificity of an elevated body temperature system. The data we provide in this report show it is possible that many, if not most, non-contact thermometry and thermographic devices are inadequate for their intended uses and investigation is urgently needed.