AI-enabled diagnosis using facial imaging
Volume 9, Issue 23 | June 13, 2024
Also in This Issue
Ancient Egypt’s pregnancy test - it’s accurate!
Home tests for COVID & flu - finally
A new way to test for HPV-positive cervical cancer
Bird-flu update
New and Noteworthy
COVID / flu tests come home (finally)
COVID brought diagnostic testing definitively into the home. By the end of the pandemic, we estimated that five to seven times more tests were being done at home vs. in the lab. The expectation was that the next big infectious-disease test to move into the home was influenza: Flu A and Flu B. (As it turns out, according to the Washington Post, the next big things in home testing didn’t have to do with infectious disease at all - they’re tests for cancer, hard-to-diagnose diseases, and “lifestyle” issues.)
The first EUA for an at-home COVID/ flu test was a LAMP-based test issued back in March 2023, but it was all alone for a year. Now the floodgates are starting to open. In the last two months, six companies have received EUAs for their OTC COVID / flu tests. All of them are lateral-flow tests that use nasal swabs as a sample.
Commentary: Given that most of COVID / flu OTC authorizations occurred after the 2023 - 24 flu season, there is no useful data yet on consumer uptake. We are cautiously optimistic, but certainly not certain.
Interest in COVID testing peaked when lab testing took too long to deliver results. Now that COVID is endemic and COVID isolation guidelines have all but disappeared, will most consumers care to know whether they have flu vs. COVID?
On one hand, knowing what is making you sick brings comfort to most people and should inform precautions to be taken. On the other hand, if you have to pay out of pocket for a test, do you need the clarity if the treatment is the same (i.e., rest and fluids)? It may be that the main market for these tests is folks who are at high risk of getting very sick from COVID or flu or live in communal and/ or multi-generational housing.
With AI, the Dx is as plain as the nose on your face (maybe)
An experienced human can diagnose a lot just by looking at a patient. Peter Rosen, who wrote the first textbook on emergency medicine, taught this skill in the waiting room at what was then Denver General Hospital. He would send students there and then ask them to describe each patient’s primary complaint and differential diagnosis based on observation alone. Two recent reports show how AI-enabled analysis is beginning to do (sort of) the same thing, looking only at the face.
The first AI-enabled tool was evaluated as part of a student’s PhD thesis at the University of Gothenburg, Sweden. It used short videos (½ to one minute long) to assess vital signs in COVID-era emergency-department patients. Using both visible and infrared light, it was able to measure blood pressure, pulse, respiratory rate, oxygen saturation, and temperature, though not as consistently as clinicians could.
The second tool evaluated patients who were suspected of having coronary artery disease (CAD; disease of the arteries within the heart). Using infrared thermal (IRT) camera images, the tool measured the temperature of various parts of the face, how those temperatures compared to one another, and how they varied over time. According to results published in BMJ Health and Care Informatics, it was about 40% better than the typical clinical scoring systems at determining whether a given patient had CAD or not. (In the area-under-the-curve chart shown here, PTP is a current clinical scoring system, and IRT is the AI-enhanced infrared thermal system.)
Two important caveats for this tool. All 460 patients (who were measured for both AI model training and evaluation) had been referred by their primary clinicians for suspected CAD. In addition, the authors did not further examine the clinical implications (false positives/positive predictive value or false negatives/negative predictive value).
Commentary: AI’s superpower is its ability to extract relevant clinical output (the signal) from data that’s diverse and highly variable (full of noise). So far, it’s been applied almost exclusively to advanced diagnostic procedures, particularly ones that require imaging. Using AI in the context of novel, less invasive, lower-cost sample types and instruments is the future of this field.
A new and effective cervical cancer test for HPV-positive folks
Being infected with certain types of human papillomavirus (HPV) is a major risk factor for cervical cancer. Right now, the next step for a patient who’s diagnosed with one of those types of HPV is cytological analysis, in which a pathologist examines cells from the cervix under a microscope. That might change in the future, based on results of a study published recently in Nature Medicine.
The research compared cytological analysis to a molecular test called WID-qCIN. Instead of evaluating what the cells from the cervix look like, the test detects whether certain genes within those cells are active or not. And it works. The combination of HPV testing plus WID-qCIN detected 100% of invasive cervical cancers and 93% of serious precancerous lesions. It was also able to predict 69% of all cancers and precancerous lesions up to six years after the sample was taken (a big improvement over the current 18%). And the study wasn’t small. It involved over 28,000 women ages 30 and up, nearly 2,400 of which had HPV 16 or 18 (the bad ones).
Commentary: The study was done in Stockholm with mostly ethnic Swedes, so the test will need to be validated in more diverse populations. Assuming it holds up, it could be a game-changer.
Bird flu update: Add Wyoming to list; how to ramp up testing?
Wyoming’s first case of H5N1 in dairy cattle was reported last Friday. It’s the 12th state to report the presence of the disease in cows.
A KFF Health News Article covered the low level of H5N1 testing in humans and how it might be increased. Only 40 people had been tested for the disease at the time of publication (June 11). Experts recommend using existing tests for influenza A (of which H5N1 is a subtype) to test anyone with symptoms. That’s not being done now because insurers typically won’t cover flu tests outside of flu season. A government mandate for coverage would change that. They also recommended that the FDA allow flu tests to be run on eye swabs, since one of the known cases in humans was only picked up on a sample from the eye - swabs from the nose and throat came back negative.
Food for Thought
Diagnostic testing has been around a lot longer than you think.
Need a pregnancy test but can’t find one at the pharmacy? Grab a bag of barley or wheat at the feed store instead. According to a 3,000-year-old medical text from Egypt’s Middle Kingdom, “Let the woman water them daily with her urine,” and if she is pregnant, the seeds will grow. If she isn’t, they won’t.
And guess what? It works - probably because the extra estrogen in pregnant women’s urine acts as a growth stimulant.