Pixel luminance artifact measurement, calibration, and correction for accurate thermographic body temperature determination

Beall, Erik B.

doi:10.1117/12.2597044

Cited by 1 publication

(5 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…3a resulted in F = 0.24. This has been revised to F = 0.1595 using the datasets collected in this study and described in Sec 3.1, all of which were collected after the implementation of a SSE correction{10} as shown in Fig. 3b, which shows inner canthi surface measurements, oral thermometry and corrected body temperatures, with the body correction derived from the linear fit of the offset between oral thermometry and surface temperature to the environmental temperature.…”

Section: Rationale For the Test Methodsmentioning

confidence: 99%

“…The second factor relates to 1) the spatial coverage of a pixel on a surface with varying temperatures and 2) the dependence of a pixel’s signal level on those around it{10}. Regarding spatial coverage, a given pixel maps to an area, or pixel spot, of a surface it is focused upon and the size of this pixel spot is based on the optics, resolution and geometry between the pixel and target.…”

Section: Introductionmentioning

confidence: 99%

“…NCITs aimed at the forehead or temporal artery from distances greater than a few mm are particularly confounded by this issue, as can be readily demonstrated by placing an NCIT into surface temperature mode, finding a particularly warm location of the forehead by repeat measurements and then acquiring temperatures of this hotspot while increasing distance. The spatial variation of skin temperature across the forehead presents a serious concern that NCITs in real-world usage are unlikely to reliably acquire a local maximum{10} and furthermore, are unlikely to capture said maximum with sufficient single-pixel coverage. While this can be observed with NCITs in surface temperature mode, it is somehow not observed in body temperature mode (possibly due to undocumented normalization of the data, which we discuss later).…”

Section: Introductionmentioning

confidence: 99%

“…This confound is known in the thermal sensing and imaging field as size-of-source effect (SSE){11,12,13,14} but is not assessed in the IRT febrile screening standard{7}. A recent conference publication{10} showed this can cause inner canthi surface temperatures to vary by an order of magnitude more than the surface calibration indicated. The effect can be described as “cooler pixels close enough to warmer pixels will pull the warmer pixels down”, so colder nose, cheeks or forehead regions (or the wearing of a face mask or hat) can cause inner canthi temperatures to read lower by as much as a degree or more, despite this effect being too subtle to visualize from a thermal image.…”

Section: Introductionmentioning

confidence: 99%

“…The effect can be described as “cooler pixels close enough to warmer pixels will pull the warmer pixels down”, so colder nose, cheeks or forehead regions (or the wearing of a face mask or hat) can cause inner canthi temperatures to read lower by as much as a degree or more, despite this effect being too subtle to visualize from a thermal image. A simple test method to quantify this effect was demonstrated{10, 15}, but to our knowledge, no image-based system currently discloses performance on SSE.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Principles and test methods of non-contact body thermometry

Beall¹,

Askegaard

Berkesch

et al. 2022

Preprint

View full text Add to dashboard Cite

Significance: Far infrared (IR) has a long history in thermometry and febrile screening. Concerns have been raised recently over the accuracy of non-contact body thermometry. Clinical testing with febrile individuals constitutes the standard performance assessment. This is challenging to replicate, which may have inadvertently allowed approval of IR systems that are unable to detect fevers. The ability to test performance without relying on febrile participants would have ramifications for public health, especially if this discovered undisclosed differences in accuracy in widely used devices. Aim: Identify foundational issues in, demonstrate principles of, and develop test methods for non-contact body thermometry. Approach: We review foundational literature and identify confounds impeding performance of IR thermography (IRT) and non-contact IR thermometry (NCIT) for febrile screening and demonstrate corrections for their effects, which would otherwise be unacceptable. Almost none of the devices we are aware of compensate for these confounds. We reverse-engineer surface-to-body temperature relations for several FDA-cleared NCITs. We note their similarity to recently reported bias-to-normal behavior in other devices and determine range of body temperatures for which the device would produce a "normal" (non-febrile) output. Finally, we generate predictable elevated face temperatures in healthy subjects and demonstrate this in several devices. Results: The surface-to-body relationships for two IRT and one NCIT were linear, while all others exhibited nonlinear bias-to-normal behavior that produce normal temperatures when presented with surface temperatures ranging from hypothermia to moderate-to-severe fever. The test method was used in healthy, non-febrile subjects to generate elevated temperatures corresponding to body temperatures from 97.35F to 102.45F. Three out of five systems had negligible sensitivity. Conclusions: This demonstrates an alternative evaluation method without the limitations and risks of febrile patients. These results indicate many devices may be unusable for body thermometry and may be providing a false sense of security for public health surveillance.

show abstract

Section: Rationale For the Test Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%