China and South Korea both take people’s temperatures (via non-contact thermometers) in a broad variety of situations. Is testing someone for a fever a useful diagnostic in isolation, without the other parts of their operational workflows, i.e. reverse GPS tracking for contact tracing?
If so, it may be useful to research or brainstorm methods of taking a person’s temperature that do not require an actual thermometer. Thermometers appear to be in short supply on Amazon, but it looks like you’d have no trouble getting one on eBay. That said, there maybe not be enough thermometers in stock to implement a testing program where every single place of business has one (or more) to test all of their employees and customers.
My first idea was to use a smartphone camera and attempt to extract some signal from that, but it turns out that not only do many smartphone cameras come with infrared filters to improve image quality, even those that don’t still aren’t capable of detecting thermal IR.
Can anyone think of any other plausible side-channels by which you might estimate a person’s body temperature within sufficient tolerances?
I had this idea below and pitched it to OpenAI—they said “”we looked into this and dont think we can do a great job with it :(”—but perhaps people here might be interested to explore it further.
Idea for zero marginal cost, digital thermometer to help contain coronavirus:
Heart rate can be estimated via (webcam or smartphone) video of someone’s face with high accuracy (even with poor video quality).[1],[2]
This heart rate might then be used to detect fever[3] (perhaps even to estimate core temperature).[4] priors such as demographic data could be used to aid detection. For instance, mean heart rate over an hour of +80 in young healthy men seems to be a robust predictor of fever.3
Fever (body temperature ≥38°C) is the most typical symptom of C19 - in 88% of confirmed cases.[5] (Though some C19 transmission might be asymptomatic[6] and presymptomatic.[7],[8])
A smartphone or web app (ala donottouchyourface.com) could be a digital fever thermometer. A webcam could continuously monitor people’s temperature and alert them to it if they have a fever (might detect anomalous increases in heart rate).
‘Thermometer Guns’ have drawbacks: they’re more expensive, you need to get close to someone’s head to take temperature, they are not very accurate, they don’t provide continuous measurement- yet it is still used for coronavirus containment.[9]
This might be a very cost-effective intervention to diagnose coronavirus.
Audio could be recorded to detect dry cough.[10], [11]
Can Smart Thermometers Track the Spread of the Coronavirus?
Non-EEG Dataset for Assessment of Neurological Status v1.0.0
[1] “Detecting Pulse from Head Motions in Video—People.csail.mit ….” http://people.csail.mit.edu/balakg/pulsefromheadmotion.html. Accessed 18 Mar. 2020.
[2] “Heart rate estimation using facial video: A review—ScienceDirect.” https://www.sciencedirect.com/science/article/abs/pii/S1746809417301362. Accessed 18 Mar. 2020.
[3] “Fever and Cardiac Rhythm | JAMA Internal Medicine | JAMA ….” https://sci-hub.tw/https://jamanetwork.com/journals/jamainternalmedicine/fullarticle/606966. Accessed 18 Mar. 2020.
[4] “Real-time core body temperature estimation from heart … - NCBI.” 13 May. 2015, https://www.ncbi.nlm.nih.gov/pubmed/25967760. Accessed 18 Mar. 2020.
[5] “Report of the WHO-China Joint Mission on Coronavirus ….” https://www.who.int/docs/default-source/coronaviruse/who-china-joint-mission-on-covid-19-final-report.pdf. Accessed 18 Mar. 2020.
[6] “Presumed Asymptomatic Carrier Transmission of COVID-19 ….” 21 Feb. 2020, https://jamanetwork.com/journals/jama/fullarticle/2762028. Accessed 18 Mar. 2020.
[7] “Potential Presymptomatic Transmission of SARS-CoV … - NCBI.” https://www.ncbi.nlm.nih.gov/pubmed/32091386. Accessed 18 Mar. 2020.
[8] “Transmission interval estimates suggest pre-symptomatic ….” 6 Mar. 2020, https://www.medrxiv.org/content/10.1101/2020.03.03.20029983v1. Accessed 18 Mar. 2020.
[9] “‘Thermometer Guns’ on Coronavirus Front Lines Are ….” 14 Feb. 2020, https://www.nytimes.com/2020/02/14/business/coronavirus-temperature-sensor-guns.html. Accessed 18 Mar. 2020.
[10] “A Cough-Based Algorithm for Automatic Diagnosis of … - NCBI.” 1 Sep. 2016, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5008773/. Accessed 18 Mar. 2020.
[11] “Cough Sounds | SpringerLink.” https://link.springer.com/chapter/10.1007/978-3-319-71824-8_15. Accessed 18 Mar. 2020.
That’s false. The accuracy isn’t high. I learned from the last conversation I had with EA who had a startup that did this, that the accuracy isn’t high enough to be useful medically. I’ll send you the contact in a message given that it’s likely who you want to talk to when you want to persue this further.
Where do those ≥38°C come from? From what I read the Chinese are using 37.3°C as a cut of for medical decision making with COVID-19.
Interesting data point—there are several papers on this that say it’s a reliable way to measure heart rate (less than 10bpm; see “Heart rate estimation using facial video”). Perhaps this could be brought down much further by throwing more engineering brains, computation and priors at it.
I saw this number in some places—for instance:
https://www.who.int/csr/disease/coronavirus_infections/InterimRevisedSurveillanceRecommendations_nCoVinfection_03Dec12.pdf
https://www.nejm.org/doi/full/10.1056/NEJMc2003100
But perhaps your number is better (source: https://www.who.int/csr/disease/coronavirus_infections/InterimRevisedSurveillanceRecommendations_nCoVinfection_03Dec12.pdf ).
I think there might be non-trivial differences due to time of the day and ethnicity as well.
The first paper you cite for measuring heart rate is of such a low quality that it didn’t pass peer review. They had only 18 subjects, did PCR and did their prediction on their trainings data.
Table8 in Heart rate estimation using facial video suggests that all of the reviewed studies had a mean error that was higher then the 5bpm that the authors call an acceptable error margin in a dynamic scenario.
The first paper that I cite has a very illustrative video and is a seminal paper in this field.
Table 8 in the review paper that you refer to shows a trend of estimation techniques getting better over time. In the latest study from 5 years ago the mean error was down to 6.47.
My broader point is:
the error rate might be brought down even further by better methods, video quality, and priors
this might so that it a valid proxy for fever
This might be very cost-effective on a population level, given the zero marginal cost of software
However, I do agree that this is not trivial.
Most commercial infrared camera’s that are used for industrial purposes aren’t accurate for measuring temperature with a high enough accuracy for measuring human body temperature. I think you do need a hardware solution and can’t bootstrap this on smartphone sensors.
We can use smartphone sensors to look at people’s throat and do machine learning. I think that has some potential to provide for a new easy to distribute tool and I’m writing a post for LW about that.
That said thermometers are quite cheap and the price of temperature sensors is cheap enough that many devices have them. It might be possible to build more thermometers out of chips that look differently then our normal thermometers if the production of normal thermometers is at capacity.