Unindicated
The Cost of Looking for Trouble
There is a spa coming to San Francisco that will scan your entire body in sixty seconds. The renderings show hot tubs, saunas, cold plunges, and rooms bathed in soft golden light, and somewhere among these comforts you will recline in a shallow pool ringed by roughly half a million ultrasonic transducers, each the size of a grain of fine sand, while sound waves assemble a three-dimensional map of everything inside you. Midjourney, the company best known for conjuring images out of text prompts, announced this venture in mid-June and named it, with a straight face, Ultrasonic CT. Imaging, the company promises, that is “as powerful as MRI and as casual as a trip to the spa.” The scans, it says, are almost a side effect. You barely think of them. And then suddenly you possess a vast library of data about your own health.1
If the phrase is going to enter the language, we may as well coin it properly. What Midjourney describes is computed tomosonography, a cross-sectional picture of the body built not from X-rays but from echoes. The hardware is real enough; the ultrasound modules are licensed from an established firm, Butterfly Network, which felt obliged to issue its own statement clarifying its role after the announcement made the technology sound like a fever dream of generative software.2 But Midjourney Medical does not yet exist as a product anyone can buy. It has no regulatory clearance, no peer-reviewed validation, and a flagship location penciled in for late 2027. It is a prototype and a press release. What it makes vivid, though, is a phenomenon that is already here, already taking your money, and already generating litigation.
That phenomenon is the elective whole-body scan sold to people who feel perfectly well. Its most prominent vendor is Prenuvo, which offers a radiation-free whole-body MRI read by affiliated radiologists and marketed as capable of detecting “500+ conditions across 33 organs.” Memberships run from roughly twelve hundred dollars a year to several thousand, the company has performed well over a hundred thousand scans, and it enjoys the kind of celebrity endorsement that money can buy, including a famous social-media post calling the machine “lifesaving.”3 Prenuvo’s strongest piece of evidence, a study it calls Polaris, reports that among about a thousand mostly asymptomatic adults, 2.2 percent were found to have a cancer confirmed by biopsy. It is worth knowing that Polaris has been presented only as a conference abstract, not published in a peer-reviewed journal, that it was conducted at a single site without a control group, and that it measures detection rather than benefit. It reports no effect on how long anyone lived.4
Here the public divides into two camps that regard each other with mutual incomprehension. On one side stand the technologists and much of the lay public, who hold a simple and intuitive conviction: more information about your body can only help you. They marshal stories, and the stories are genuinely moving, of the friend whose tumor was caught early on a scan done for some unrelated reason, the colleague who would be dead but for an incidental finding. On the other side stand most physicians, who hold a proposition that sounds, to the first camp, almost perverse: that bad data can be worse than no data at all. The doctors worry about incidental findings, the ambiguous shadows and indeterminate nodules that lead to biopsies, to surgeries, to a cascade of investigation that carries its own real harms. The two sides are not weighing the same things. One counts the lives a scan might catch. The other counts the harms a scan reliably causes. And anecdote, by its nature, only ever reaches the first ledger.
I will confess that my own sympathies are not entirely where my white coat would put them. I am excited by this technology. The vision the enthusiasts describe, a machine that can look inside a living body and reliably sort what matters from what does not, is a thing I want to exist, and I believe the only road to it runs through exactly the kind of restless private invention these companies embody. Nothing that follows is an argument that we should not build such a device, or that the dream is foolish. It is a very good dream. But the dream is not what is being sold. What Prenuvo sells today, and what Midjourney proposes to sell tomorrow, is at best the same MRI a clinician would order for a patient with an actual complaint, and often something less capable. The picture is no better than what medicine already has. By the one measure that matters it is worse, because it arrives stripped of the clinical question that gives any image its meaning. The enthusiast imagines he is buying an early glimpse of the perfected scanner. He is buying an ordinary scanner used in the single circumstance guaranteed to degrade what it can tell him. These companies trade on the credibility of a future product to move a present one that is, by that measure, a step backward.
I want to be honest about how hard this is, because the hardest case I know is my own family. My mother, who is herself a physician, had a renal cell carcinoma found incidentally on a CT scan ordered for an attack of diverticulitis. The tumor was about a centimeter across. For a solid renal mass that small, the guidelines do not tell you to operate. They tell you to watch it, because a tumor under four centimeters rarely behaves aggressively, and because the risks of surgery, the bleeding, the loss of kidney function, the operative complications, generally outweigh the benefit of removing something that was very unlikely ever to harm you.5 The evidence here is not soft. A decade of prospective data on small renal masses shows no meaningful difference in cancer-specific survival between the patients who had surgery and the patients who were simply observed.6 My mother knew all of this. She had the tumor removed anyway, largely at her own insistence, because she could not abide the thought of a cancer sitting inside her, and she knew she would never relax again if she chose to do nothing. As her son, I find it almost impossible to argue with her. As a physician trying to be dispassionate, I have to admit that what she chose was, by the evidence, the wrong call.
That contradiction is the whole matter in miniature. A trained doctor, presented with her own data, did not behave as a calm Bayesian weighing risks and benefits. She behaved as a person who had just learned she had cancer. And if a physician cannot reliably hold that line, it is fantasy to imagine that the technology executive, or the worried-well thirty-something with disposable income, will hold it either. This is the fact the enthusiasts keep stepping around. The problem with an incidental finding is not only that it is often a false alarm. The problem is that once you know about it, you cannot un-know it, and the doing-nothing that the evidence recommends becomes, for almost everyone, psychologically unavailable. The scan does not just give you information. It gives you a burden you did not have to carry, and then it sells you the surgery to put it down.
Why should bad data be worse than none? The answer is a matter of arithmetic that every clinician understands and most patients are never taught. The usefulness of any test depends on how likely the disease was before you tested. Run a scan on a population riddled with a disease and even an imperfect test is informative. Run the same scan on a crowd of well people, where almost no one has the thing you are hunting, and the math inverts: most of the alarms it raises will be false, and a great many of the rest will be findings that were never going to matter. This is not a hypothetical worry. A systematic review of whole-body MRI in asymptomatic people found a pooled false-positive rate around one in six, with indeterminate or critical incidental findings common and no study able to demonstrate that any of it prolonged a life.7 Prenuvo itself reports that roughly half the people it scans walk away with something to keep an eye on. And the consequences of all that watching are documented. In a national survey of physicians, virtually every respondent, 99 percent of them, reported having personally seen an incidental finding set off a cascade of further testing, and they described those cascades inflicting psychological harm, physical harm, and financial harm on their patients.8 The look is cheap. The looking-into is not.
If you want to know where this road ends, it has already been traveled, and the country that traveled it ran the experiment cleanly enough to publish the results. Beginning in the 1990s, South Korea added cheap thyroid ultrasound to its health screening, and physicians began finding tiny thyroid nodules in enormous numbers. By 2011 the rate of thyroid-cancer diagnosis was fifteen times what it had been in 1993. It looked like an epidemic. It was not. Thyroid-cancer mortality over the same period did not budge, the unmistakable signature of overdiagnosis, of finding cancers that were never going to kill anyone.9 What the screening did change was the number of thyroids removed. Most of those patients underwent surgery, and surgery has a price that does not appear on the brochure: in an analysis of more than fifteen thousand of them, 11 percent were left with hypoparathyroidism and 2 percent with vocal-cord paralysis, and most now depend on thyroid hormone replacement for the rest of their lives.10 Then came the part that proves the case. A group of physicians publicly urged the country to stop screening healthy people, and the country largely did. Diagnoses fell by about a third. Operations fell by more than a third. And nobody was worse off, because the cancers that vanished from the statistics were the ones that were never going to harm anyone in the first place.11 You cannot ask for a more direct demonstration. Screening asymptomatic people manufactured disease, the disease was treated with morbid surgery, and the cure was simply to stop looking.
There is a further problem, specific to imaging, that the technologists have not reckoned with, and it concerns the radiologist whose name goes on the report. A scan ordered for a reason comes with a question attached. The patient has chest pain, or a palpable mass, or an abnormal blood test, and that clinical context tells the radiologist what he is looking for and, just as importantly, what he can safely ignore. A whole-body scan on a well person arrives with no question at all. It asks the reader to evaluate everything and to privilege nothing, which is not a richer task but an impossible one, because significance is most often not a property an image has on its own. It is a judgment that depends on context the scan has deliberately stripped away. Confronted with that, and mindful of his own exposure, the careful radiologist does the only safe thing. He flags everything he possibly can. The over-calling is not incompetence but the rational response of a person who knows that the finding he dismisses is the one that will be read aloud in a courtroom.
And the courtroom is a very real consideration. A New York man who paid twenty-five hundred dollars for a Prenuvo scan in 2023, and was told his brain looked normal, suffered a catastrophic stroke some months later in that very region. His attorneys, citing a neurologist they retained, contend that the original images showed a sixty-percent narrowing of a major cerebral artery that the report failed to mention, and that it might have been treated had anyone documented it.12 The case is winding through the New York courts, and I would not pretend to know how it comes out.13 But notice the bind, because it cuts in both directions at once. The radiologist who misses something is sued, as this one was. The radiologist who over-calls everything generates the cascades we have just spent this essay cataloguing. There is no safe reading of a scan that was performed without a reason, because the absence of a reason is precisely what makes a confident read impossible. That impossibility is not a failure of the radiologist. It is the predictable product of pointing a camera at a healthy man and asking it to find his future.
It is worth being precise about which findings actually cause the trouble, because the enthusiasts are not entirely wrong about some of them. The lung nodule and the small thyroid mass, the staples of the overdiagnosis literature, are in truth the manageable case. We know how to watch a nodule. We follow it on serial images, most are never biopsied, and a device that could track such a thing cheaply and without radiation, should it ever perform as promised, would be a real convenience, whatever it did to the raw count of procedures. The heart of the problem lies elsewhere, in the finding we do not know how to act upon at all. Return to that alleged sixty-percent narrowing of a cerebral artery, and suppose it was really there. What follows from knowing it? No surgeon opens the skull to repair a moderate, nonobstructive atherosclerotic lesion in an otherwise well man. The finding does not change his treatment. It changes one thing only: it gives him, and his lawyer, someone to blame when the stroke eventually comes. And where such findings do prompt action, the action is invasive. An incidental coronary calcification can send a patient to a cardiac catheterization; a suspicious cerebral vessel can send him to catheter angiography of the brain. These are not biopsies, but they carry their own real risks. Even the gentler follow-ups are not free, because the confirmatory CT or the catheter study delivers the very radiation these scans are advertised to spare you, and the contrast that CT and MRI often require carries hazards of its own. The point bites hardest against the ultrasound spa, because sound waves penetrate the body poorly, and the vague signals such a device produces will almost always have to be chased with the more capable modalities, the CT and the MRI, that it was meant to render unnecessary. The radiation-free scan turns out to be the first step on the road to the radiation.
The enthusiasts have an answer ready for all of this, and the answer is artificial intelligence. The machines will read the scans, they say, and the machines will not tire or err or hedge. I am skeptical, and the reason follows directly from everything above. An algorithm has no more clinical context than the human radiologist does, because the context is missing from the scan itself, not from the reader. The model still has to decide what counts as a finding and what does not, still has to separate the shadow that matters from the thousand that do not, and it must make that call, as the human must, without the clinical question that would tell it which is which. Significance is contextual, and a system handed no context cannot supply it. AI does not dissolve the radiologist’s predicament. It automates it, and at the scale Midjourney imagines, a billion scans a month, it automates it across the entire population at once.
There is a second reason to doubt the machine, and it concerns not what the model can do but how it would have to learn. The tech-minded reply to everything above is scale: feed the system millions of scans and it will learn, as we did, to tell the ominous from the harmless. But a scan does not arrive labeled. The model cannot learn significance from images alone, any more than a radiologist could, because significance is established only by what the workup eventually proves. The label on the training image is the biopsy result, the surgical pathology, the catheter that confirmed or refuted the lesion. To teach the algorithm what matters, in other words, hundreds of thousands of people must first undergo the invasive workup that produces the answer. The training data is human morbidity. Perhaps that bargain is even worth striking, in the long run. But it is a fair one only if the people being scanned understand that they are not merely customers. They are the training set, and the tuition is paid in their own biopsies.
Some online defenders wave all of this away with a single move: if the incidental findings are so troublesome, just ignore them. The suggestion barely survives contact with how medicine works. A physician cannot be shown a finding and then unsee it, and a patient cannot be told there is a shadow on his scan and then be expected to forget it. The law will not permit the pretense from either of them. We have spent this entire essay on the fact that knowledge, once acquired, cannot be set down at will. “Just ignore it” is not a policy. It is a wish.
So what should a free society do about a product that healthy people want to buy and that the evidence says will mostly harm them? The libertarian reflex, and I share its starting point, is to say it is my body and my money and my business. Liberty is the governing principle, and I do not reach for prohibition lightly. But look closely at what is actually happening here, because it is not a free market functioning. It is a market failure wearing the costume of consumer choice. Two distortions give the game away. The first is that the buyer does not bear the true cost of his purchase. The scan is paid out of pocket, but the cascade it sets off, the confirmatory MRI, the biopsy, the specialist referral, the operation, is billed to insurance and spread across everyone in the risk pool. The purchaser keeps the reassurance and the novelty. The rest of us get the invoice. That is the textbook definition of an externality, and a market in which the buyer captures the benefit while strangers absorb the cost will always produce too much of the thing. The second distortion is that the liability lands on the wrong people entirely. The radiologist who never wanted to read an unindicated scan carries the malpractice exposure, and so does the internist downstream who did not order the thing and must now chase its incidentalomas to ground. The company that marketed the scan to a well man, that built the spa and ran the advertisement and took the fee, walks between the raindrops.
The remedy is not the prohibitionist instinct, the impulse to forbid the product outright and treat grown adults as wards of the state. The remedy is to make the market honest, which means making the prices and the liabilities tell the truth. Let people buy the scan. But let the seller bear what the scan sets in motion. If a company sells you a look, that company should own the workup the look generates, or insure against it, so that the downstream cost is priced into the ticket rather than mailed to the insurance pool. Aim the tort liability at the party that created the risk and profited from it, the marketer who solicited a healthy customer, rather than at the contracted radiologist or the treating physician left holding the bag. And require the same honest disclosure at the point of sale that we would demand of anyone else making a health claim: the false-positive rate, the rate of incidental findings, and the plain fact that no whole-body scan of asymptomatic people has ever been shown to make them live longer.14 This is what tort reform should mean in this context. Not a shield for the companies selling the scans, but liability aimed correctly, at the people who sell them.
What we have instead is the worst of both arrangements, the aesthetics of a free market joined to the economics of a subsidized one. A company forced to price in the strokes it fails to predict and the thyroids it needlessly removes would have to ask whether its product survives at its true cost, and many of these scans would not survive the question. That is not a verdict against the future. I began by confessing that I want the machine the enthusiasts describe, and I meant it. The day will likely come when a device can look inside a well person and tell him, reliably, what is worth his worry and what is not, and that day will be an advance worth celebrating. But it arrives through honest validation, honest prices, and an honest assignment of who answers for the consequences, not through a spa that sells a worse scan today on the credit of a better one promised for tomorrow. My mother’s choice should stay with us here, because it was not irrational and it was not ignorant. It was human. Once a person knows there is a cancer inside him, he will rarely choose to do nothing, and no amount of counseling reliably changes that. It is the permanent condition of being mortal and aware of it. Which is exactly why the looking should not be sold casually, as an amenity beside the cold plunge, to people who arrived with no question to answer and no way to bear the answer they are given. The honest price of a look includes everything that follows from it. Sell it that way, to people whose eyes are open, and you have a market and perhaps, in time, a genuine advance in medicine. Sell it as it is sold now, and you have a sales pitch, with the bill addressed to the rest of us.
Midjourney. A New Era of Midjourney: Announcing Midjourney Medical. June 18, 2026. https://www.midjourney.com/medical/blogpost
Claburn T. Midjourney pivots from AI image generation to body-scanning medical spa where patients bathe in “golden light.” The Register. June 18, 2026. https://www.theregister.com/ai-and-ml/2026/06/18/midjourney-pivots-from-ai-image-generation-to-body-scanning-medical-spa-where-patients-bathe-in-golden-light/
Stempniak M. Kim Kardashian promotes whole-body MRI service from Prenuvo, drawing concern from some. Radiology Business. August 10, 2023. https://radiologybusiness.com/topics/healthcare-management/healthcare-economics/kim-kardashian-promotes-whole-body-mri-service-prenuvo-drawing-concern-some
Westgate C, Nayeri R, Datta M, Venstrom J, Pompa RS, Shingru P, Basar S, Hashemi S, Chodakiewitz Y. Abstract 7406: Noncontrast screening whole body MRI with diffusion-weighted imaging for multi cancer detection: a retrospective case series study. Cancer Research. 2025;85(8_Suppl_1):7406. https://aacrjournals.org/cancerres/article/85/8_Supplement_1/7406/760042/Abstract-7406-Noncontrast-screening-whole-body-MRI
Campbell SC, Uzzo RG, Allaf ME, et al. Renal Mass and Localized Renal Cancer: AUA Guideline. Journal of Urology. 2017;198(3):520-529. doi:10.1016/j.juro.2017.04.100. https://www.auajournals.org/doi/10.1016/j.juro.2017.04.100
Pierorazio PM, Johnson MH, Ball MW, et al. Five-Year Analysis of a Multi-institutional Prospective Clinical Trial of Delayed Intervention and Surveillance for Small Renal Masses: The DISSRM Registry. European Urology. 2015;68(3):408-415. doi:10.1016/j.eururo.2015.02.001. https://www.europeanurology.com/article/S0302-2838(15)00121-9/fulltext
Kwee RM, Kwee TC. Whole-body MRI for preventive health screening: A systematic review of the literature. Journal of Magnetic Resonance Imaging. 2019;50(5):1489-1503. doi:10.1002/jmri.26736. https://onlinelibrary.wiley.com/doi/10.1002/jmri.26736
Ganguli I, Simpkin AL, Lupo C, et al. Cascades of Care After Incidental Findings in a US National Survey of Physicians. JAMA Network Open. 2019;2(10):e1913325. doi:10.1001/jamanetworkopen.2019.13325. https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2752597
Ahn HS, Kim HJ, Welch HG. Korea’s Thyroid-Cancer “Epidemic” — Screening and Overdiagnosis. New England Journal of Medicine. 2014;371(19):1765-1767. doi:10.1056/NEJMp1409841. https://www.nejm.org/doi/full/10.1056/NEJMp1409841
Ahn HS, Kim HJ, Welch HG. Korea’s Thyroid-Cancer “Epidemic” — Screening and Overdiagnosis. New England Journal of Medicine. 2014;371(19):1765-1767. doi:10.1056/NEJMp1409841. https://www.nejm.org/doi/full/10.1056/NEJMp1409841
Ahn HS, Welch HG. South Korea’s Thyroid-Cancer “Epidemic” — Turning the Tide. New England Journal of Medicine. 2015;373(24):2389-2390. doi:10.1056/NEJMc1507622. https://www.nejm.org/doi/full/10.1056/NEJMc1507622
Whole-body MRI provider Prenuvo loses bid to limit damages in high-profile malpractice case. Radiology Business. January 14, 2026. https://radiologybusiness.com/topics/healthcare-management/legal-news/whole-body-mri-provider-prenuvo-loses-bid-limit-damages-high-profile-malpractice-case
Prenuvo’s full body MRI missed signs of a catastrophic stroke, lawsuit says. The Washington Post. January 13, 2026. https://www.washingtonpost.com/health/2026/01/13/prenuvo-lawsuit-full-body-scan/
American College of Radiology. ACR Statement on Screening Total Body MRI. April 17, 2023. https://www.acr.org/News-and-Publications/Media-Center/2023/ACR-Statement-on-Screening-Total-Body-MRI