You’re probably familiar with one tell-tale sign of a respiratory illness – a rough, throaty, distressed voice that can tell you when someone you’re close to is unwell (or, conversely, needs to be carefully mimicked when you call your employer to say “I…*cough*… don’t think I can come in today”).
But evidently, our voices may hold more clues about our health than just how our lungs are working, and technology developers, clinicians and AI researchers are trying to identify “biomarkers” — those tell-tale signs – in our voices to help identify potential health concerns such as Covid-19, but also notably non-respiratory conditions including dementia and heart disease.
Analysis of our speech, made possible by the ubiquity of mobile devices and the number of anonymized and randomized samples they can provide, may offer opportunities for differential diagnosis – whether you in fact have a cold or just cheered too loudly at your child’s hockey game – and could help with the triage of patients, identification of potentially infectious carriers, or even home diagnosis of a number of ailments.
AI and machine learning are at the heart of this, supported by smartphone technology and crowdsourcing – machine learning works best when the machines are given many opportunities. AI has already shown itself proficient at analyzing ECG (electrocardiograph) data as well as clinicians, however ECGs are specialized diagnostic methods requiring a trained technician and dedicated medical equipment.
Analysis of the speech of individuals both with and without COPD (Chronic Obstructive Pulmonary Disease) led to the development of an app which can help clinicians identify patients at risk, and could possibly do the same for Covid-19 and other illnesses. Having lungs full of fluid is audible to anyone and requires no machine learning, but what special sounds do illnesses like depression or cardiac anomalies make?
This technology is still young, but a 90% accuracy rate identifying PTSD (Post traumatic stress disorder) or an 80% accuracy rate in identifying autism – in infants under one year of age! — hint at future potential for this technology (and also warn us of impending privacy concerns.).
If, a few years from now, your employer insists that requests for sick leave be made via telephone rather than e-mail, you’ll know why. You heard it here first.
To read more: Alexa, do I have COVID-19? (nature.com)
———-
Artificial intelligence examining ecgs predicts irregular heartbeat, death risk. (n.d.). Retrieved April 01, 2021, from https://newsroom.heart.org/news/artificial-intelligence-examining-ecgs-predicts-irregular-heartbeat-death-risk
Anthes, E. (2020, September 30). Alexa, do I Have COVID-19? Retrieved April 01, 2021, from https://www.nature.com/articles/d41586-020-02732-4
Thanks for this, Sean –
You focus is on voice, which is huge, but we additionally reveal ourselves in an extraordinary number of other ways that can be analyzed insightfully via machine learning. Just a few examples:
– our breath and body odors (sensors in elevators, work places, etc.)
– our excretions (smart toilets and sewer systems)
– our facial expressions (do you think you can lie to a machine that knows how to look?)
– our body movements (every person can be uniquely identified by their walking gate, for example)
– our biorhythms (heart rate, blood pressure, skin temperature, brainwaves, etc.)
– our DNA and microbiome
– our behaviour (manners, habits, etc.)
All of these are ‘available’ to external analysis. And yes, we purposefully analyze most of these already in human-to-human interactions (we ‘spy’ on other people), without the aid of computers, so we shouldn’t believe precedents haven’t been set. I would be deeply scared about these futures in an authoritarian system, and we need to exceptionally vigilant about it, but I’m also inspired by opportunities here for personal development through insights I might gain about myself that might otherwise be unavailable to me. The key is my security and autonomy in relation to all of these data.
One thing that makes me optimistic, beyond my delusional belief in human nature, is that most of the data can be mediated through my smartphone (whatever that device looks like in the future), so as long as I can exert some kind of legal control over how it employs my data, it can be beneficial for me. For example, with your example of voice, if my smartphone detects some malady that I wish to hide, it could modulate my outgoing voice on-the-fly so that no external analysis could detect that malady. Also, if I wanted a sick day, I could leave a message that was entirely convincing.
Lots of Black Mirror episode potentials here…
This is outstanding technology! Wow, thank you so much for sharing! Very interesting immediately made me think of the medical bay robots in the movie Elysium which imediatly diagnosis the patient by scanning them. This is obviously different but the concept of immediate diagnosis is apparent in both. I’m my opinion unless it’s 100% percent accurate I don’t see it as a deciding factor for a teacher sick day. These are my reasons
1. Most teachers are part of a union and unless it’s 100% it would be a nightmare for the school district if they forced an employee to work who actually was sick. There could be legal implications of workers’ rights and not to mention the remedy. This would be especially detrimental in a situation like COVID if the teachers who called in sick infected others.
2. Mental health days are becoming more common and in many districts. By the time this technology is ready for use, both will most likely fall under the same category of sick days. Unless the phone would also 100% diagnose the mental state of an employee it wouldn’t really work.
This is defiantly an eye-opener! Thanks again for sharing.
Hi Michael. I think you’re probably right — I doubt we’ll ever be asked to phone the Sick Day hotline and recite a test phrase — but it might be possible that some day this technology would work in the opposite direction, proving that you are sick. Below, I alluded to core temperature being a familiar bio-marker that can indicate infection. I doubt many employers would tell a worker with a normal temperature that they cannot be sick, but neither could I imagine an employer telling a worker with a fever of 39.4C/103F that they’re not.
Hi Sean. This is amazing technology and a bit scary to think that our voices can be used as a source of information. That being said, I was thinking about what happens when I call in sick to our dispatch line. I’m not sick very often, but when I call in I don’t try to disguise how rough my voice is and I’ve had the thought in the past that I didn’t sound sick enough to be calling in. The other end of that is that many educators will go to work when they are sick (not so much in 2020-2021) because it’s often more work to prepare for a TOC than it is to just drag yourself to school. This kind of diagnostic technology could be used in work sites in the future to make sure that workers are healthy enough to be there.
Hi Lori. It certainly is amazing, and I’m sure that further research will uncover new bio-markers (voice or otherwise) for other conditions, and that are similarly easy to check and and analyze, but I suspect their use will be limited to triage and early detection more so than definitive diagnosis. I’m sure we’re all currently familiar with the use of IR thermometers to screen the public for fever (another biomarker than correlates with potential Covid infection). It’s not especially accurate — plenty of false positives and false negatives — but it’s cheap, non-invasive and takes just seconds, so it’s a good “coarse filter” and I suspect that other bio-markers might be used similarly.
The simplicity of obtaining a measurement does give pause for some concern, from a privacy point of view. In a different post on hacking your cellphone I facetiously suggested making your phone into a thermal imaging camera to see who in your class is sick. I was joking, of course, but since we can’t control our own bio-markers, it wasn’t necessarily unrealistic.