This isn’t earth-shattering stuff, but I do think it’s a tidy illustration of how a small, easy-to-make change with a relatively minimal amount of hassle can nonetheless reap a small but measurable benefit—and in the long term, meaningful time and energy savings.

I appreciate/hope that this will all be irrelevant for radiologists very soon, but while thousands of us are still using Powerscribe, this is still part of our worklife.
Read More →

A few weeks ago, I got to enjoy a pitch for another poorly conceived “revolutionary” radiology AI workflow and reporting tool.

Tech people: Just bring on a radiologist CMO and a couple more to work on product. Stop cheaping out. Give them stock if you can’t pay, but these mostly suck and will continue to suck.

Everyone is happy to have some radiologists as “partners” that are just customers beta testing your buggy, half-baked products for free, but not enough are using content experts to make useful software from the get-go using first principles rooted in real-world experience and expertise. I would also love to see less focus on peddling trash and more on building product.

Free advice: Maybe just build something straightforward using current capabilities that is easy to deploy integrate into current workflows that people want right now. Something that doesn’t require massive buy-in and changing your whole tech stack.

Enterprise software sucks. Build up some good will, go from there. Not everyone needs to raise a ton of money to milk the bubble of me-too “AI for X” wrappers. Make something that solves a small, specific, real pain point and enjoy a nice cash-flowing business for a few years.

Reinvest in the next product if you want—or don’t. Forget about multiple rounds of raising capital trying to build and scale a behemoth on a foundation of sand.

Now, if you really want to revolutionize everything and replace radiologists with magical AI powers, great, that’s totally fine. You may be able to skip lots of radiologist feedback (though I imagine you’d still be better off with some deeply integrated, thoughtful radiologists). Someone somewhere can revolutionize everything from farm to table, but there’s also low-hanging fruit to optimize specific parts of the workflow in the meantime. Every part of the imaging pipeline has tedious, essentially broken software tasks and inefficiencies, and in many situations, it’ll be easier to optimize them individually in the short term than try to replace everything wholesale.

In other news, if you’re a current software vendor, now is the time to improve your offerings before it’s too late.

Everyone is happy to play the enterprise software game and court big hospital systems. But no one wants to build a grassroots business working with real people doing real work—because it doesn’t scale easily and it’s hard to raise money for.

I know that getting customers is hard—but that may be because your product sucks, because of the friction involved in transitioning to an unproven solution, or because you can’t demonstrate real benefits beyond just saying “AI.”

Yes, inertia is real: your new thing needs to be way better than the incumbent or something you can plug in for a reasonable additional cost. That still leaves a lot of opportunity on the table.

It’s job-hunting season, and I’ve received a variation of the following question several times this week alone: “How do I figure out if a practice I interview at might sell to private equity?”

I appreciate the fear of joining a private practice only to have the rug pulled during the workup in a sale to private equity. It’s what I was worried about when looking for jobs in 2017, what I was scared of when I entered practice in 2018, and what happened to some of my friends in 2018-2019.

My group was and is fiercely independent, and I was fortunate that the Dallas area was not ripe pickings for Radiology Partners, unlike Houston and Austin markets. But I had many of my friends end up on the wrong end of a sale and eventually change practices.

I have also certainly spilled enough digital ink on this topic over the years myself, so I am probably not entirely free from blame for increasing the collective anxiety about this issue.

But I do think that at this juncture, it’s relatively low risk.

The era of PE expansion in radiology through debt-fueled acquisitions of individual practices is essentially over, as far as I can tell. This is a model almost entirely dependent on the zero interest rate environment of the twenty-teens. The costs of borrowing money now are too high to enable these shenanigans, and the degree of leverage these companies have is already so high that there really isn’t any excess capital to deploy in acquiring individual practices when they also need to service their debt, pay for operations, and invest in AI and other magic.

Furthermore, the PR is not great at this point, and I doubt most practices that are actually healthy would want to sell. No one is buying the initial magic & sparkles pitches, so I don’t think either party wins in 2025, and everyone knows it. A struggling practice wanting to hitch their ride to a larger organization and/or extract some value before implosion would be a different story—but those would be less desirable for a purchase. RP, USRS, and LucidHealth may not be that good at actually running a radiology business, but they are very good at their real business, which is a primarily finance game that happens to involve healthcare.

So, investing tens of millions of dollars (even if you had them to burn) in an individual practice acquisition is very risky in 2025. Since these companies have reached scale, there are better ways for them to grow.

Private equity is more likely to grow their workforce through hiring individual radiologists than they are through group purchases, and they’re more likely to grow their imaging volume through organic growth or contract sniping than they are through the outright purchase of a practice. They can also grow by picking up the pieces when someone else fails, like RP did when Envision “transitioned” the corpse of its radiology business.

The “hostile takeover” is still somewhat possible, in the sense that an RP or similar could swoop in and try to steal a contract from a local group, have that local group dissolve because that contract represented a large fraction of their business, and then hope to hire up some of those radiologists for free on the back end to essentially keep the jobs they already had but have since lost (as in, keep staffing the hospitals they were already staffing before the contract change).

This has happened before, but even this, I think, is relatively unlikely to happen now or happen at scale, because these PE companies are not immune from the challenges in the market and have a hard time staffing as well (and also because many hospitals aren’t particularly happy with their level of service).

The reality is that private equity hasn’t gone away and won’t go away, but the greater fear for an individual practice is to implode under the weight of unsustainable image volume growth or be unable to provide the right lifestyle and compensation balance that are required to hire and retain radiologists in this increasingly nationwide market in the era of teleradiology.

A group failing because they can’t be competitive in the job market because their hospital won’t pay for the stipends to make their job competitive, for example, is a real concern. Could a PE-entity swoop in and hoover up some work there? Absolutely, but that’s not the same thing as your new practice screwing you over.

This is to say: If the job sounds good enough that you want to do it, then I personally wouldn’t worry much about it at this point. A healthy group probably doesn’t have much to fear from private equity in the short term given the radiologist shortage. The market itself is enough of a challenge.

When I was a fellow, my key metric when choosing my job was variety, not so much in terms of pathology or the pictures themselves but in the day-to-day. Variety helps me do one overarching critical thing for my professional satisfaction: optimize for enthusiasm.

There are a lot of things I like about radiology and some that I don’t, but one thing that makes everything go down smoother is a nice balance to the week—with different kinds of work on different kinds of days that demand different kinds of things from you.

Academia

I always thought I would be in typical academic practice because I generally like being “involved,” and I like the community. I enjoy teaching and mentorship, and I always have. I’ve been a peer mentor of one variety or another since high school. It’s just something that I find meaningful. I don’t didn’t even mind committee work and other kinds of bureaucracy, even if how the sausage gets made is off-putting. (If we’re being honest, I also probably felt I’d stay in academia because of comfort with the only system I’d ever known, willingness to buy into the lie that the best work gets done there, and a general failure of imagination.)

One thing I didn’t like was the rigid hierarchy, the prevailing pay-your-dues to get a better job attitude, the unfair treatment and distribution of different kinds of work among different kinds of people, especially when such treatment is a preference for seniority or clout that is sometimes unearned and often counterproductive for actual department functioning.

Another was that I don’t particularly enjoy the research game, which is the only meaningful academic currency in many departments, even when most research we do as a field—and certainly all the research I’ve done—has been trash. Someone should do it, but it doesn’t need to be me.

A bit of research here and there is fine, but a job with a “clinician educator” focus (and where that is valued) is what spoke to me.

Academia providing less vacation and money, while often true, wasn’t actually that much of a consideration at the time.

Physician-Ownership & Governance

As those who have read the relevant posts on this site, I wanted nothing to do with private equity when I found my first job. Thankfully, that wasn’t an issue in Dallas. I don’t regret that outlook, and my peers who joined PE or joined practices that sold to PE have all left without exception.

With regards to university or hospital employment, it doesn’t take much exposure to the layers of management or dubiously useful hierarchy to find a representative democracy with physicians in control to be refreshing. Obviously, the hospital is still the hospital and is dysfunctional in all the ways large organizations so often are, but a true private practice is one-step insulated and removed, able to advocate for ourselves and control our own workflows.

Speaking to hierarchy, I liked that new associates could be involved in everything and do anything in the group except serve on the board of directors. I was an associate program director for the residency before I was a partner.

A Truly Hybrid Schedule

My schedule is a combination of teaching at the hospital, working from home, and sometimes working solo at an imaging center in a strip mall somewhere.

Ultimately, I’m a better teacher when teaching is part of my job that I get to look forward to and not something I do every single day. Trust me, even I sometimes get tired of hearing myself speak and saying the same things over and over again.

I also like the fact that I get to spend some of my days working with residents and students, and some of my days working by myself—reading my own cases at my own pace, sometimes doing my own procedures and talking (briefly) to my own patients.

(I did just take over the program director role this month, and that also means regular admin time as well.)

Having a hybrid schedule was especially important for me, because while I have no interest in being a teleradiologist, I very much have an interest in working from home on at least a weekly basis (I’m about 50% remote).

My wife has her own solo psychiatry practice and makes her own schedule. So as the parents of two young children, the ability to have lunch together or take a walk around the neighborhood during the week is an incredible boon to our marriage. We’ve had more mini dates and spent more quality time together during the day than we could ever hope to carve out from our busy evenings with the family or over-scheduled weekends.

Some flexibility is seriously valuable.

A Four-Day Workweek

I also really appreciated my group’s goal of a four-day workweek. I didn’t really need an academic day—which can get filled by meetings and duties and other administrative tasks—what I really wanted was a day off to pursue my hobbies/interests, to be a good partner and parent, take my kids to school, pick them up, make sure the house is in order, and yes, recharge my battery at least just a little.

Some groups have lots of week-long vacation blocks because it is by far the easiest way to do scheduling, but the reality is that I don’t need week-long blocks in huge numbers. In my current life, I’m not going on lots of trips when my kids are in school. I also don’t need large blocks of time off because I’m not diligent enough in my time management to take advantage of them for creative pursuits—I need regular time off.

So a four-day workweek combined with healthy vacation is a great mix. When you combine my eight weeks of PTO and my four-day workweek, you end up with the equivalent of like 17 weeks off, which is, of course, phenomenal.

From a partnership perspective, this is also a strong way to handle staffing in that the fifth day off is not a guarantee depending on who’s on vacation and how the lists are, but it allows for our division to be flexible accomodating shifting workloads, scheduling PTO even one day at a time, and staffing/recruiting.

This is to say, I don’t always get that fifth day off—but I do most of the time. And when I don’t, I just get paid extra for working.

But when people ask how I have time to write, that’s certainly part of the answer. 1I wrote this last Thursday while sipping my Cometeer iced coffee (p.s. help me, I’m running out).

//

My situation is, of course, mine. My wife became an attending when I was a senior resident, and we’d already returned to be near our families for residency. We weren’t going anywhere. I wasn’t canvassing the country looking for the most magical of all possible jobs. I was looking for the best job for me in the Dallas area, and I was trying to achieve that combination of work variety, location variety, and schedule flexibility.

There’s nothing wrong with different kinds of radiology practice. Different strokes and all that. But I will admit that when I was a trainee deciding between staying in academics at my home institution and joining a traditional, typical private practice, I couldn’t shake the feeling that I wasn’t excited about either choice/extreme. I feel very fortunate that the chance to do something in the middle was available where I was looking.

(And yes, we’re hiring.)

In his 2023 book Decisions about Decisions, Harvard Law School professor Cass Sunstein offers this advice: Rather than concentrating on the probability of being right or wrong in a decision—which is often impossible to determine due to the intrinsic uncertainty and the unpredictability of the future—focus instead on comparing the cost of being wrong with the benefit of being right. These factors, according to Sunstein, are easier to estimate without the need for forecasting outcomes.

Applying this very logical argument to using high-quality AI tools for diagnostic medicine, we come to a straightforward problem as fleshy, fallible humans: the logical course of action is to agree with any plausible AI answer and only contradict the machine in cases of undebatable error. This is true for potential liability, but it’s also true just for saving face generally.

If the goal is to maximize accuracy or quality, one can imagine a world where a human radiologist interprets a scan independently and an algorithm interprets a scan independently. If both agree, then we’re done. If those two evaluations are in disagreement, then a third party—either another human or a different algorithm with different parameters—steps in to adjudicate the disagreement. (We could, of course, have that initial AI product itself be the result of a debate between multiple algorithms, but you get the idea.)

There is no guarantee that such a combination would be an improvement, but it’s a plausible outcome that will, of course, be studied. However, the effectiveness of such an approach remains uncertain. How much would such a system genuinely enhance diagnostic accuracy? Surely, it would be a moving target, but would such human-AI collaboration genuinely enhance accuracy or would it be awkward source of complexity and hamstring the needed efficiency gains. It certainly wouldn’t look very good if the third party nearly always sided with one source.

Potential Outcomes of AI-Human Collaboration

There are several possible outcomes:

1. The human is usually right, and the addition of the AI does not create a significant change.
2. The AI is usually right, and the addition of a human does not create a significant change.
3. The human is usually right, but the AI results helps catch what would be unequivocal bone-headed mistakes.
4. Both the human and the AI are usually right, but in cases where they disagree, a third-party adjudicator adds additional value by catching edge cases with higher frequency than either individual alone. If nothing else, the third party creates the system that is needed to handle discordance.
5. Alternatively, the combination could result in overall reduced accuracy. For example, the AI is almost always right, but the uncertainty of human disagreement actually reduces the overall accuracy.

That will be studied. Yet, reality could be complex—we may find that AI’s strengths and weaknesses differ across imaging modalities, patient populations, or specific pathologies. AI may be great at breast screening but terrible at most MRI. Or the opposite. The optimal balance between AI independence and human oversight may depend on more or different variables than we’d suspect. Or not. Why pretend to know?

The Likely Commercial Model for AI in Radiology

The commercial reality is that the sort of AI utilization I just described is unlikely to be the primary solution for handling the radiologist shortage or maximizing profitability for stakeholders unless it’s a rule. The more likely scenario is that AIs will churn out preliminary reports of increasingly high quality, which a human radiologist will review, make changes to, and ultimately be liable for.

This shifts the radiologist’s role from a thoughtful creator and analyst to more of a quality inspector—checking for plausibility rather than deeply analyzing every case. When the AI is reasonable, the human will likely agree. When the AI makes an obvious mistake, correcting it won’t require much effort from the human. Obvious contrast mixing in a pulmonary artery is not a thrombus. Calcified lymph nodes are often chronic findings, etc. A clearly benign breast lesion misclassified as a potentially malignant tumor may be easy for an experienced mammographer to catch, especially if that mammographer has access to priors and context that the AI does not.

It’s easy for many observers with a vested interest to believe that their magical subset of skills will be particularly thorny to emulate, and some may even be right.

Even the quality inspector assumption presumes a relatively stable and predictable level of AI performance. How confident should a human be in their assessment when there is disagreement if the AI is improving while the human is mostly stagnant? What if AI-generated reports vary significantly in quality for different use cases? Scrutiny may be hard to employ judiciously in a piecemeal fashion.

Regulatory agencies could impose strict requirements for human oversight that make the process more labor-intensive than expected, and those requirements could be either reasonable or stupid over the short, intermediate, and long term. AI adoption will depend not only on technical feasibility but also on evolving legal, ethical, and financial pressures.

The Risk of Automation Bias

But what will radiologists do when the AI calls a focal asymmetry that the radiologist would not have called? We’re getting there already. If the AI is usually right, the human being will almost certainly just agree with whatever it says as long as it’s plausible—because the risks of agreeing are negligible, but the risks of incorrectly disagreeing are high.

How foolish will you feel calling a mammogram normal when the AI suspects a mass—with its black-box, pixel-based approach that detects patterns beyond and different from your human understanding? No one wants to get in the way, so no one will disagree and take on the liability of calling a case negative when the AI has flagged it as positive in an otherwise usually accurate system.

That’s the reality we need to live in. That’s what we’re going to see unless we specifically craft one to prevent it.

That’s going to be a big problem—because all the commercial and workforce pressure will push us toward utilizing AI tools in ways that practically ensure automation bias becomes the single biggest challenge facing radiologists in the near future.