An app can transform smartphones into thermometers that accurately detect fevers | Hacker News Reader

An app can transform smartphones into thermometers that accurately detect fevers

washington.edu

74 pointsgeox54 comments

> it uses the phone’s touchscreen and repurposes the existing battery temperature sensors to gather data that a machine learning model uses to estimate people’s core body temperatures

> FeverPhone estimated patient core body temperatures with an average error of about 0.41 degrees Fahrenheit (0.23 degrees Celsius), which is in the clinically acceptable range of 0.5 C

> with an average error

Eww, that's not how it works. The goal isn't to make the average near the limit, since the average of -100 and 100 means 0 average error! The goal is to make sure the measurement have a distribution that gives some reasonable confidence that it's within the limit.

This is PTT: https://www.spcpress.com/pdf/DJW244.pdf

I would have naively calculated average error as avg(max( abs(smallest), largest)). so it would be 100 in your case. I'm sure that's not statistically correct but it makes sense to me.

Abstract says "mean absolute error of 0.229 °C"

It’d be hard to get average error so low, since you need precision to get low error values for varying temps.

You can’t just return 100 and -100 randomly, since the actual temp is unknown and you’d need to evenly sample around that.

A normal distribution will average to 0, with enough samples. The center and spread is important, since individual measurements can be considered a random sampling within that distribution. But, this is overly simplifying things, and assume random error, which AI systems often don't exhibit.

It appears this is moot, because the quote did not reflect the actual value or method used.

The distribution of human temperature levels is already known, you could just give the average temp as every guess and get maybe 1.5 F average error.

My point is, you can’t really get 0.44F avg error unless it’s predictive. The “std dev” of abs(reality - model) is too high

Skimming through the study, it seems like most of the control temps were within a 0.6° C range. I wonder what the average absolute error would be of a thermometer that always returned 37°C

According to a sibling comment the abstract claims "a mean absolute error of 0.229 °C". According to wiki the normal temperature window is 36.5–37.5 °C and that of a fever is 37.2-37.7 °C or higher, with variations for both of about 0.5 °C throughout a 24 hour period.

> An early morning temperature higher than 37.2 °C (99.0 °F) or a late afternoon temperature higher than 37.7 °C (99.9 °F) is normally considered a fever, assuming that the temperature is elevated due to a change in the hypothalamus's setpoint.[14] Lower thresholds are sometimes appropriate for elderly people.[14] The normal daily temperature variation is typically 0.5 °C (0.90 °F), but can be greater among people recovering from a fever.

https://en.wikipedia.org/wiki/Human_body_temperature

Amazing!

As a parent to a toddler I was surprised to note how bad I am at estimating whether me or my child have a fever.

On top of that infants naturally can have a body temperature of up to 37.5°C, which is an additional hurdle because they may feel warm, but be generally fine.

Years of soldering taught me that the lips are very sensitive to temperature(much more so than fingers) and heal fast should you touch something really hot, like a soldering iron, so for emergencies there's that, but obviously it doesn't work when both of you have a temperature.

Also, for more than one reason, you can't reliably check yourself.

Check under this chin/side of the neck. It's more stable then the forehead which changes temperature frequently.

I hope you’re using lead-free solder if you’re holding it with your mouth. The fumes aren’t great at that distance either

Why on earth is your soldering iron anywhere close to your mouth?!

Temperature check

Surely you don't need something as sensitive as your lips to detect if something is ~270c?

Yes and no actually. Fingers work, but interestingly the sensation of heat comes with enough delay to cause burns before you register the temperature.

My soldering iron had barely 15W, so it would frequently get too cold.

Also I could avoid smooching the thing for the most part, because the heat radiating from it was already noticeable from a few millimeters - particularly due to moisture evaporating from the lips at an increased rate.

I replaced that iron with a more powerful one which didn't have this issue, but I still use this method if I'm unsure what's the temperature of something that might well be very hot.

> interestingly the sensation of heat comes with enough delay to cause burns before you register the temperature.

As someone who solders pretty regularly: don't I know it!

> Also I could avoid smooching the thing for the most part, because the heat radiating from it was already noticeable from a few millimeters

Totally works with your fingertips too. If it's "too cold" to feel without touching, it's cold enough to give a quick tap, which'll inform you about it's temperature some more. But I guess everyone has their own tricks!

I've switched to a TS-100 a while back, and having a temperature readout right on your soldering iron is really quite nice.

I often used the touch temperature of my children tights. Somehow this matched the "fever or not" threshold shown by the thermometer when I finally got home of it.

Now that the children are teenagers/ young adults I find plenty of thermometers all over the place.

When we had our newborn we were surprised to find that many of the expensive infrared thermometers can be quite inaccurate compared to a simple $10 probe thermometer you stick under an armpit.

My mother-in-law is a nurse and they all use the cheap Omron probe thermometers.

When I had Covid, my Omron mouth thermometer was reading 103F while every infrared thermometer read my forehead at 99F.

I don't use infrared thermometers anymore.

I'd have started by measuring the inside of your mouth with the infrared, personally.

Most doctors ask for an anal probe temp for newborns before they ask you to bring them in. The armpit method gets significantly harder when they get older, more clothes, fussy, etc. I got a Braun Thermoscan 7 and it’s worth every penny. Makes it super straight forward to get an accurate temp when they’re sleeping. Also since it has a disposable sleeve, you don’t have to wash the thermometer after every use like under the tongue. You can keep using the same one the whole length of the sickness.

I asked my doctor about this when our first was born. He said it’s a baby not a soufflé. An approximate temperature is all we need to get a sense of the general trend of a fever.

Our cheap Braun ear thermometer has been wonderful even if it’s not super precise.

Sometimes I think there’s a ton of theatre in modern medicine.

> Sometimes I think there’s a ton of theatre in modern medicine.

It's a free market, baby!

If you have to take temperature several times over a week of moderate illness, an approximate measure is enough.

When your baby has 104F temperature in the middle of the night and you are deciding whether to take them to the ER or not, that's when you wish you had forked over for a more expensive and accurate model.

If you’re on that precipice, you should just be going.

ERs have their own risks, particularly catching something new. We tried our best to keep our 26 weekers out of it as much as possible, as even a cold was pretty disastrous in the first couple of years.

I doubt the data supports this mindset (unless there’s some specific pre-existing issue that puts someone at elevated risk). Everyone makes their own choices in trying to be the best parents they can be, and I respect that.

I just don’t think “expensive anal thermometer to protect kids from dangerous hospitals” passes the sniff test for me. I’ll stick with my Ivermectin and Q-ring bracelet.

Assuming you were talking about a Q Ray bracelet, there doesn’t appear to be evidence of it working: https://www.psychologytoday.com/us/blog/homo-consumericus/20...

> I doubt the data supports this mindset

Doubt all you like, it's a thing. "In the United States, the Centers for Disease Control and Prevention estimated that roughly 1.7 million healthcare-associated infections, from all types of microorganisms, including bacteria and fungi combined, cause or contribute to 99,000 deaths each year."

ERs are dirty, busy places filled with pathogens. It's not uncommon to share the waiting room or a semi-private room with other patients with unknown conditions throwing up or coughing, with the resulting droplets/areosols in the shared air. Even before COVID, going to one when it's a borderline situation requires a bit of calculus, especially if you're medically complex.

https://en.wikipedia.org/wiki/Hospital-acquired_infection

> unless there’s some specific pre-existing issue

That'd be the 26 weeker thing I mentioned, yes.

Your comment reads like you’re talking about an anal thermometer, and I was wondering how in the world that doesn’t wake up a sleeping baby. It looks like the Thermoscan 7 is an ear thermometer though, which makes a lot more sense!

It’s got disposable tips and you are only assuming he is indeed using it in ear! I have the same one. I do now wonder if it would work, but I’m not going to try it.

I am pretty skeptical how often it comes back at 98.6 vs 98.5 or 98.7, they have a deadband in there I think.

Is it anal or rectal? I’d always heard the latter.

You use a rectal thermometer to take an anal probe.

Either word is fine. There's no need to be rectal about it.

We’ve tried every thermometer under the sun, and all but the probe models start out inaccurate or quickly crap out. Infrared, temporal probe, etc. One of my biggest parenting pet peeves. We now only do the under-the-tongue method, and only if the good ol’ palm-to-the-forehead thermometer says that something is off. I read a study somewhere that indicated that parents are actually pretty accurate when using that method, so I’m sticking with it, replicable or not.

"The devices can range from $15 to $300, and many people need them only a few times a year."

What kind of thermometers do you need in the USA to measure a common fever??

They are less than $5 in EU [0], but I'm sure it's because of our "higher taxes" and "long waits".

[0] https://www.amazon.it/Pic-Family-Termometro-Digitale-Pezzo/d...

The last thermometer I bought here in the US was $1.25. However, it is still true that thermometers can range from $15 to $300. And a smartphone app that solves a $300 problem sounds a lot more impressive in a press release than one that solves a $1.25 problem.

This isn't an economics article.

Interesting concept. No download links though, the app doesn't seem to exist anywhere yet.

At least in the apple ecosystem, I learned that any medical claim must be backed up by clinical studies.

My simple heart rate monitoring app was almost blocked for not having any studies, and I had to point to the dozens of other apps that simply report the heart rate from an external sensor.

Also, I doubt there’s a public API for getting the battery temperature.

In the paper (6.4 Limitations), there is an explanation for it:

..."Lastly, FeverPhone uses root-level access for reading a broad set of thermistors and a custom user-debug Android kernel for accessing raw touchscreen capacitance data. These requirements limit FeverPhone to phones with fine-grained sensor access. Temperature data from the battery thermistor is more readily accessible but did not yield comparable performance relative to temperature data from other thermistors. Still, future work could explore a more accessible version of FeverPhone with specific considerations paid to the battery thermistor. Similarly, the TouchEventAPI is publicly accessible at the cost of coarse-grained information about touch location. The use of this API in place of the raw capacitance data could be explored in future work."

Sure many here and elsewhere run on rooted Androids.. My case at least. I'd love to try it.

> a custom user-debug Android kernel for accessing raw touchscreen capacitance data.

I don't know much about android and kernels, but seems like only rooting your phone still would not be suffice.

Hmm, Cornell also has a smartphone app called FeverPhone[1] which appears to have quite different functionality...

1: https://techcrunch.com/2016/06/21/nih-grants-2-3m-to-develop...

crazy how many functions smart phones have and to see new ones develop is great

Apple should definitely include stuff like this in their iPhone hardware. (an actual thermometer). They have the premium and margin to do so and it would make the iPhone even more sticky as a platform.

Thermometer and other biometrics would be great. In theory you could measure a pulse by touching the phone and it detects the vibrations, runs that through some ML. Maybe possible already?

Isn't it on the iWatch?

Yes but not for the wearer to take measurements of their body temperature for example to see if they might have a fever. It only takes readings at night while the wearer is asleep, and only provides data in the form of trends against your individual baseline.

So it can be useful to see if you were generally hotter last night than normal, which could be useful when sick, but you can't open a thermometer app and check your actual temperature which would have more utility.

The pulse is probably doable but difficult, at least on an iPhone. The noise in the accelerometer data from just holding it is pretty substantial.

Pulse is pretty trivial with another sensor on pretty much every smart phone out there: the camera

I've been using pulse apps for probably at least a decade

I’ve compared iPhone pulse apps to my Polartec H10 chest strap and they are wildly different many times. More than 10% difference easily.

I wonder if it still works when the phone is really hot, or if that could interfere with the sensor?

Reply on news.ycombinator.com