Author Topic: Thermal Image dataset for Fever and Human Temperature Measurement  (Read 2083 times)

0 Members and 1 Guest are viewing this topic.

Offline eri0o

  • Contributor
  • Posts: 15
  • Country: br
Hello,

Are there any known datasets of images of humans with and without fever, along with true measurements of their fever?

Regards.
 


Offline Bill W

  • Frequent Contributor
  • **
  • Posts: 499
  • Country: gb
    • Fire TICS
Re: Thermal Image dataset for Fever and Human Temperature Measurement
« Reply #2 on: March 25, 2020, 03:20:57 pm »
Not seen any, and I doubt a scientific approach is taken especially at present.  Many think cameras are just to make it look good and perhaps pick up a few people, and that may be a worthwhile approach.

Ideally you'd let 100 people walk past the camera, and see who it picked up.
However you must also check the ones it did not pick up, and then compare camera to measured body temperatures.

Someone who has just run upstairs or is nervous might be unusually hot (false positive, tedious but safe)
Someone with a fever who has just been outside might look OK due to surface cooling (false negative, very bad).
I found I tend to look cold, not hot, when getting ill.

As we know cameras only see surface temperatures, and the airport scanning types are only landing a few pixels on someones' head.  The only reliable check for fever is to get at their internal temperature.

Bill

Offline eri0o

  • Contributor
  • Posts: 15
  • Country: br
Re: Thermal Image dataset for Fever and Human Temperature Measurement
« Reply #3 on: March 25, 2020, 03:49:39 pm »
I am just looking for the dataset. While I agree with the topics you mentioned, the truth is I would like to look into a dataset myself, and see what can be done with it.

Unfortunately finding thermal images in general that are classified and properly noted is already hard, in the specific case of what I am looking into I could not find any. Still, I wonder if there's some medical or thermal database that is just not found by Google that exists and some person wandering in these forums has access to.
 

Offline DaJMasta

  • Super Contributor
  • ***
  • Posts: 1901
  • Country: us
    • medpants.com
Re: Thermal Image dataset for Fever and Human Temperature Measurement
« Reply #4 on: March 25, 2020, 08:56:54 pm »
There could be, but I think generally groups taking and using such datasets would likely publish their findings and not their data - even if in low resolution, it's also likely just publishing a collection of pictures of the public, which does bother some.


It would be quite easy to take one of your own, but people are likely sensitive to having a camera (or a weird pistol camera industrial thing) pointed at them without an explanation.  Given that the base accuracy of most thermal cameras isn't actually great, to get a reasonably high accuracy in picking out fevers you probably need a calibration reference either in frame or periodically in view, then a situation where hair/head coverings/angle and distance to the target are somewhat constrained (an interior hallway after a coat room, for example) to get reliable data.


That said, you could probably just take a picture of a crowd and pick out a couple of outliers regardless of the actual measured value and with a short investigation (maybe another picture with their mouth open if they haven't drank anything immediately before or just a check to see if they're sweating) could probably pick out fevers in a large group pretty reliably.  I've actually been using my thermal cam to take pictures of my face for the past week or so to get an idea of what the response looks like, and it seems to be the best representation of body temperature on the face is either side of the tongue with your mouth open - but if you take a sip of water before measurement, it drops a fair bit.  Otherwise, environmental conditions can lead to more than a full degree of variation in the face, even when at a fixed distance/focus and with a subject not wearing headclothes or long hair - but if comparing against others in the same environment, you could likely compensate for the absolute variation by comparative measurement.
 

Offline Stray Electron

  • Frequent Contributor
  • **
  • Posts: 972
Re: Thermal Image dataset for Fever and Human Temperature Measurement
« Reply #5 on: March 25, 2020, 09:16:51 pm »
   There have been a few pictures published of the Chinese and others using thermal cameras to screen passengers at airports. I wonder if any of the pictures show displays with readable numbers?  Fraser started a thread on that subject several weeks ago.
 

Online Fraser

  • Super Contributor
  • ***
  • Posts: 9441
  • Country: gb
 
The following users thanked this post: edavid

Online Fraser

  • Super Contributor
  • ***
  • Posts: 9441
  • Country: gb
Re: Thermal Image dataset for Fever and Human Temperature Measurement
« Reply #7 on: March 26, 2020, 12:06:36 am »
From my above referenced thread....

Use Sci-Hub to look at this paper as it contains data that may be of interest to you.

https://doi.org/10.1118/1.1819532

Fraser
 

Online Fraser

  • Super Contributor
  • ***
  • Posts: 9441
  • Country: gb
Re: Thermal Image dataset for Fever and Human Temperature Measurement
« Reply #8 on: March 26, 2020, 12:22:27 am »
 

Offline eri0o

  • Contributor
  • Posts: 15
  • Country: br
Re: Thermal Image dataset for Fever and Human Temperature Measurement
« Reply #9 on: March 26, 2020, 03:10:39 am »
Thanks for the Papers Fraser! They are indeed interesting!

DaJMasta, do you have any images stored or tabular data on the different regions of the face as you tested?
« Last Edit: March 26, 2020, 03:12:28 am by eri0o »
 

Offline Vipitis

  • Frequent Contributor
  • **
  • Posts: 666
  • Country: de
  • aspiring thermal photography enthusiast
Re: Thermal Image dataset for Fever and Human Temperature Measurement
« Reply #10 on: March 26, 2020, 03:12:37 am »
are you looking for a labeled dataset to train a model?

That would be ambitious, but I believe isotherms and alarms can do it much easier and should be a fairly steep learning curve as it gets weighted very quick to only care about max readings.
 

Offline eri0o

  • Contributor
  • Posts: 15
  • Country: br
Re: Thermal Image dataset for Fever and Human Temperature Measurement
« Reply #11 on: March 26, 2020, 03:16:03 am »
I really really wanted train dataset...
 
But yes, what I found so far is something similar to this video, where it marks a region for screening and picks up the max temp, and buzzes an alarm if this max temp in the region is above a threshold.



(this forum could use having spoiler so the video could be hidden by default)
« Last Edit: March 26, 2020, 03:17:37 am by eri0o »
 

Offline eri0o

  • Contributor
  • Posts: 15
  • Country: br
Re: Thermal Image dataset for Fever and Human Temperature Measurement
« Reply #12 on: March 26, 2020, 01:32:06 pm »
DaJMasta , you said mouth open. A person suggested me to use the ears, specifically by focusing as possible in the auditory canal, it would be closer to internal temperature and it's a place that can be measured without disturbing people (without having them removing a mouth mask). Looks like an idea to explore.
 

Online Fraser

  • Super Contributor
  • ***
  • Posts: 9441
  • Country: gb
Re: Thermal Image dataset for Fever and Human Temperature Measurement
« Reply #13 on: March 26, 2020, 01:45:43 pm »
Eeeek ! No.... do not go down that path ! To correctly measure a humans core temperature you need to measure an area of high blood flow and positioning is all important. Even a dedicated specialist Tympanic membrane IR thermometer will provide an inaccurate reading if not positioned correctly. This is warned about in the professional Tympanic membrane type thermometer user guides. The other location that provides a descent IR thermometer measurement point is the Temporal artery as shown in the attached images. When measuring the Temporal artery, some calculation is needed to correct the reading for true core temperature !

I wrote some comment on fever measurement for a friend a while ago. I will include it in a separate post in a minute.

Fraser
« Last Edit: March 26, 2020, 02:49:26 pm by Fraser »
 
The following users thanked this post: IwuzBornanerd

Online Fraser

  • Super Contributor
  • ***
  • Posts: 9441
  • Country: gb
Re: Thermal Image dataset for Fever and Human Temperature Measurement
« Reply #14 on: March 26, 2020, 01:46:58 pm »
The comment I sent to my friend .........

Regarding testing a humans core temperature....... I know a little about that topic. The issue here is in the title..... “Core” ! Core temperature is closely controlled by the brains built in thermometer called the Hypothalamus. It’s nominal setting is around 37C and it constantly compares the bodies core temperature to that setting. If the core starts to drop below 37C, the Hypothalamus gets the body to generate heat to increase the core temperature. If the temperature of the core rises too high, it gets the skin to sweat and act as a cooling radiator. That is how it should work when a human is well.

When a human gets some illnesses, the Hypothalamus is compromised in its calibration and can cause a fever. This fever can be considered to start at a core temperature of 38C. If the persons temperature rises to 39.5C they have a high fever. At 41C things are getting really serious and the person has a very high fever. So we are seeing a change in CORE temperature of just 3C between nominal and a nasty fever. All this is going on inside the human body and is not well reflected on the exterior, non core areas of the human. In an effort to capture a more accurate non invasive core temperature measurement it is necessary to monitor an area of high fresh blood flow that has not lost too much heat getting to the outside of the body. The skin is an effective radiator so you need a big artery without too much dissipation from the skin. Medical science has used two specific locations that provide a decent core temperature monitor point. One is the the tympanic membrane in the ear as it has a good core blood supply and is well insulated in the head at the bottom of the ear canal. The other location is the true artery monitoring point called the temporal artery. It comes from the core and rises to the skins surface at the temple. A specialist thermometer is applied to this artery to measure the artery thermal output. The thermometer applies corrections and can predict core temperature if that is required. The difference between the temporal artery temperature and core temperature has been studied and is understood. The temporal artery has to be carefully located for the test to be accurate.

As you are finding the forehead is not the best indicator of core temperature. It’s thin skin coverage over the skull with relatively distributed blood flow means that the surface temperature is easily varied by external influences and a high core temperature is not always obvious unless the person is in really serious trouble. Sweat evaporating on the skin can also reduce the apparent core temperature.

So what can we tell from this ? Well unless a human has a very high temperature it can be very hard to accurately detect a true elevated core temperature. Fever cams have been proved effective when the sufferer of a fever has a very high temperature, but a slightly elevated temperature can easily be missed. A thermal camera for fever detection needs to be very carefully calibrated and configured to measure over a narrow range of temperature to detect the small temperature changes in the skin. The LUT also has to reflect the offsets applicable to Core temperature vs head temperature. As stated, it is not a great approach to determining core temperature but it is all a thermal camera can easily monitor. This is a very complex challenge. Even Temporal thermometers took a lot of development to provide accurate measurements. It remains inaccurate if the user is not trained.

So, you have a challenge on your hands if you want to create a truly effective fever cam that uses exposed skin temperature as it’s source of information.

I suspect there have been papers written on the effectiveness of fever cams but I have not seen them. I will see if I can find anything for you.

Fraser
 

Offline cuda12

  • Contributor
  • Posts: 37
  • Country: us
 

Offline DaJMasta

  • Super Contributor
  • ***
  • Posts: 1901
  • Country: us
    • medpants.com
Re: Thermal Image dataset for Fever and Human Temperature Measurement
« Reply #16 on: March 27, 2020, 03:43:12 pm »
I still think there's an opportunity to use a thermal cam to screen a crowd and pick out temperature outliers, but you have to make sure that there isn't obstructing clothing and that the individuals have been in the same conditions for some time (someone coming in out of the cold, for example, would appear an outlier until they come up to temperature with the rest of the crowd).

I took some more pics of my head with my thermal cam and in terms of maximum measurement (closest to core temperature), my best results have been in the mouth, all the way back, on either side of the tongue.  I took a couple close ups with the tongue up and down, and surprisingly, much of the underside of the tongue is still well below the temperature of those back corners between the tongue and the side of the mouth.  Of course, there is definitely a closeness requirement to get a reasonable resolution picture of your mouth which would still require protective clothing in a disease screening situation, it also doesn't really work noninvasively because they have to have their mouth wide open to be screened.  I tried a couple of pictures of the ear and it seems like it would be a real challenge to get a good reading - much harder than through the mouth.  Partly because the ear canal isn't straight so you can't really see all the way in, partly because it's tough to focus on it, and partly because it would require even closer contact with the screener.

You can see some variation day to day in relative contrast (and days where the nose isn't cold, for whatever reason) in my images, but it doesn't seem like any features on the exterior of the face are as close to core temperature as the back corners of the mouth, and with a base accuracy of +-2C, even my e40 hacked to e60 wouldn't be that reliable for absolute temperature unless you had a calibration point regularly referenced or in frame.  My images were taken indoors with very little airflow in a slightly cool environment.  Don't know if they'll be any use, but maybe as a comparison of hotspots or another datapoint to go with pictures of your own face!
 

Offline IwuzBornanerd

  • Frequent Contributor
  • **
  • Posts: 254
  • Country: us
Re: Thermal Image dataset for Fever and Human Temperature Measurement
« Reply #17 on: March 28, 2020, 09:01:55 pm »
Fraser;
Thanks much for that "map" of the arteries.  I have one of those temple thermometers and also had an ear thermometer & never read 98.xF with either of them.  ...Until I saw your images.  The instructions for the temple thermometer merely say to press the thing "firmly on the temple" with no specifics on the location, so I was always placing it back behind my eye about level with the vertical center of my eye, which puts it between the temporal & ophthalmic arteries.  I have now realized that I can actually see a ridge where my temporal artery is & when I put the thermometer on that ridge I get 98.xF.  :-+
I am not opposed to exercise, unless it is an exercise in futility.
 

Offline Ben321

  • Frequent Contributor
  • **
  • Posts: 506
Re: Thermal Image dataset for Fever and Human Temperature Measurement
« Reply #18 on: March 30, 2020, 05:24:46 am »
Interesting, but how do you know that the skin surface temperature is really accurate? I mean, normal temperature is 98.6 degF for the INSIDE of your body and is normally quite constant. Air temperature is not always the same. On a cold winter day, it might be 30 degF, or on a hot summer day it might be 100 degF.

A temperature gradient will exist from the surface of your skin into your body. The surface of your skin will be somewhere in between ambient temperature and your body temperarature. On a hot day (which is only slightly warmer than your body temperature), an LWIR scan of your forehead might say you have a fever, even if you are healthy. On a cold winder day though, your forehead surface temperature (as seen by an LWIR camera) will be something like 70 degF WELL BELOW any temperature that would be healthy for you to have for a body temperature (in fact, if that was your internal body temperature, you would be DEAD).

Forehead surface LWIR measurements are quite useless. For an accurate LWIR temperature reading, use an IR ear thermometer. That's a body cavity, so it is mostly isolated from the ambient temperature.

I would NEVER trust those coronavirus LWIR cameras setup at airports to detect fever on the foreheads of incoming travelers.
 

Offline DaJMasta

  • Super Contributor
  • ***
  • Posts: 1901
  • Country: us
    • medpants.com
Re: Thermal Image dataset for Fever and Human Temperature Measurement
« Reply #19 on: March 30, 2020, 07:52:29 pm »
I would NEVER trust those coronavirus LWIR cameras setup at airports to detect fever on the foreheads of incoming travelers.

But that's the thing, while their absolute accuracy is worse and environmental effects play a significant role, they are still very sensitive and very easy to use without contamination.  Since a screening location such as in an airport at least has a fairly uniform environment, it could still be an effective tool to narrow the scope of a search - you would still have to screen for fever using a conventional thermometer, but you could limit the number of screens and exposure of the screener by prescreening with a thermal camera and then taking the higher outliers of the crowd to actually measure with a normal thermometer.  Something like screening for the top 5% or 2% of measured temperatures on a thermal camera would probably get 90%+ of the fever cases that pass through to be tested by thermometer and it would reduce exposure to the screener, reduce the bottleneck from screening time, and reduce the exposure for the passengers because of the reduced bottleneck and less time spent waiting around in a confined space covered with public surfaces.

A thermal camera could dramatically improve the efficiency of a screening checkpoint when used in conjunction with a thermometer in most situations (the people being screened have to have their faces visible, have to be in a relatively temperature-uniform environment, and can't have just been in a different environment), even if it's difficult to pull a meaningful measurement of core temperature out of just a thermal image.
 

Offline Bill W

  • Frequent Contributor
  • **
  • Posts: 499
  • Country: gb
    • Fire TICS
Re: Thermal Image dataset for Fever and Human Temperature Measurement
« Reply #20 on: April 03, 2020, 11:41:07 am »
Something like screening for the top 5% or 2% of measured temperatures on a thermal camera would probably get 90%+ of the fever cases that pass through to be tested by thermometer

That is the big question, 90% or 10%.  While 'hot' people are measured and some have a fever, no-one knows (or wants to admit) how many 'normal' people passed by the camera had a fever anyway.

Doing thermometer testing only 'hot' people and getting some hits may be creating a false perspective that is self-reinforcing.

Bill

Online Fraser

  • Super Contributor
  • ***
  • Posts: 9441
  • Country: gb
Re: Thermal Image dataset for Fever and Human Temperature Measurement
« Reply #21 on: April 03, 2020, 12:22:48 pm »
Using a radiometric thermal camera for fever detection is still a relatively new application of the technology. It has been used to observe living targets for thermal abnormalities such as tumors and inflammation but Fever detection became a medical application for the technology during the SARS outbreak in 2002/3. It seemed perfect for the role until the human thermal characteristics were better understood. Much research followed the SARS outbreak and fever detection using thermal cameras was included in that research. It was found to be a useful tool but there several drawbacks to the technology. Those issues remain as they relate to how the Hunan body is constructed, how it manages core temperature and how it is dressed in daily life. I feel sure that further research into the best use of non contact temperature measurement of humans will follow this current Pandemic. The scientists, medics and Thermal Imaging industry are still on a learning curve where reliable fever detection is concerned. It may never be possible to use thermal imaging fever detection with the degree of accuracy that some might desire. The human body is a very complex and dynamic target where precision thermography is concerned.

What we are seeing at the moment is almost every producer of thermal imaging equipment adapting current camera technology and software to meet the needs of Governments demanding more fever screening equipment. Whether the equipment being produced by such companies is adequate for the task remains unknown at this time. A lot of medical knowledge is needed to design a thermal imaging camera and software for effective fever detection rather than just a public relations tool to make people think Governments are taking action to protect them  ;)

In years to come we may see some very specialised thermography equipment designed from the ground up for fever detection in humans from a safe distance. Such equipment may be finely tuned to that specific role rather than a generic thermal camera that is tweaked a little and cobbled together with some fancy looking software GUI.

The Temporal artery IR thermometer took a lot of development to get it right. Thermography of humans for Fever detection will likely be no different.

Fraser
« Last Edit: April 03, 2020, 04:26:31 pm by Fraser »
 

Offline eri0o

  • Contributor
  • Posts: 15
  • Country: br
 
The following users thanked this post: Fraser, Vipitis


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf