The Approaching Singularity: Part II
The Convergence of AI and Radiation Oncology
The Radiation Oncology Singularity — Part II: The Convergence
In Part I, we traced two parallel histories - artificial intelligence and radiation oncology - each crossing its own event horizon, each accelerating toward a singularity. Today, we follow those trajectories to where they land and try to evaluate their intersection.
The Original Small Circle
Today in radiation oncology, the concept is often quite simple - put dose into the disease we see on imaging. There is less “complexity” to field coverage. We use less fractions and more often, we simply put dose into the cancer and put less dose elsewhere. If you are younger, you probably appreciate this shift less. If you trained prior to 2000, it is likely more apparent.
That is what we talked about in far greater detail in Part I. I call that our singularity - the overarching vision for where nearly all of our treatment approaches are heading across nearly all sites. Old rules like 1.8 Gy - 2 Gy per fraction and anatomic coverage of “sub-clinical” regions are breaking down.
This is what happens at Singularity: the old rules break down and new ones form. And yes, the term doubly applies - both to this paradigm shift and to the visual way in which our targets and dose clouds are becoming simpler, smaller, and more concentrated.
But today, before we really get going, let’s look back at history - to the original small circle of dose.
Before 1943, thyroid cancer was a surgical disease. External radiation played a desperate and largely ineffective supporting role. The technology was basic / poor and we simply couldn’t hit the small areas of disease without causing a lot of collateral damage. Then came I-131. Developed by Dr. Saul Hertz and Dr. Samuel Seidlin - and something fundamental changed - we targeted the disease directly. It was the original small-circle delivery. It ignored the subclinical geography of the neck entirely and moved straight to the molecular target. It didn’t treat what might be there. It treated what was demonstrably, biologically present.
And the role for radiation? Radiation oncology didn’t evolve in low-risk thyroid cancer. We substituted this better targeted treatment.
Now clearly, the radiation technology in 1940 was nowhere near what it is today. The machines that allow modern treatment approaches didn’t exist. The imaging didn’t exist. And yet the logic of I-131 - find a “target” via uptake or receptor, deliver the dose, spare everything else - maps almost perfectly onto the direction our entire field has been moving for the past two decades.
Today, more and more - across a number of sites and indications, we treat visible disease. Fewer fractions, tighter margins. Put dose in the target and limit it as much as possible elsewhere - it is “our singularity”. But really, finding a unique receptor on the affected cells - flagging them and killing those specific cells clearly offers a path to better.
From the 1940s to now spans more than eighty years. Consider how much has changed. With hindsight, it makes perfect sense that the targeted approach of iodine ablation proved superior. It is after all amazingly specific - even by today’s standards. And in that context, despite all our extraordinary technological gains in external beam radiotherapy, we still choose the intravenous theranostic solution (I-131) as the more elegant and effective option for this disease.
Honestly, even from a radiation oncology perspective, it “feels” like the more intellectually satisfying approach: delivering radiation directly to the broken biology rather than to a physical volume defined by imaging.
With that bit of history, let’s look at where we are today.
Theranostics: Approaching Our Singularity From the Inside Out
Theranostics is precision medicine in its most literal form. A targeted agent identifies the cancer on diagnostic imaging. A near-identical binding agent then delivers high-energy radiation directly to those cells.
See the disease via a unique antigen.
Deliver radiation precisely via that antigen to the disease.
The subclinical geography of the surrounding tissue becomes largely irrelevant. Two modern examples that are no longer experimental:
Lutathera targets somatostatin receptors in neuroendocrine tumors.
Pluvicto targets PSMA-expressing prostate cancer cells.
These are not the blunt instrument of radium-223, which worked by exploiting high bone turnover rather than true molecular specificity. These are quite precise. Targeted. And they are working.
Now step back and look at what they are doing mechanistically. A ligand finds the lesion from the inside out. Biodistribution replaces beam geometry. The pharmaceutical supply chain replaces the capital equipment stack. The $7-10 million machine, the vault, the physicist, the dosimetrist, the scheduling infrastructure - the entire apparatus that defines a radiation oncology department - becomes, at minimum, partially redundant for the disease sites these agents address.
That is not a criticism. It is an observation about where the logic leads.
And the trials are already mapping the territory. We’ll look at Pluvicto because it has significant “overlap” with external beam approaches. Consider some trials utilizing Pluvicto: The ENZA-p trial is examining enzalutamide with or without Lu-177 in metastatic prostate cancer. BULLSEYE is exploring Lu-177 to postpone ADT. PSMAddition is looking at ADT and ARPI with or without Lu-177. None of these trials strictly prohibit SBRT in appropriate patients - but none are designed around it either. The primary pathway in each is pharmaceutical theranostics. Metastasis directed therapy with external radiation is present at the margins, not at the center.
This is not an isolated signal. It is a trend line.
Pluvicto or Radiation for Oligometastatic Prostate Cancer
And Pluvicto has been “successful” in an area where, arguably, radiation has its strongest data for benefit in the metastatic setting. Think STAMPEDE or SABR-COMET or ORIOLE/STOMP. This is one of our “homeruns” of the past decade. We have shown, rather consistently in prospective datasets, that we likely improve overall survival through metastatic directed treatment.
And yet, even here, within our strongest metastatic indication - we are often being replaced - or at least marginalized. Certainly, there are trials looking at various combinations that include SBRT, but in many - our role is relegated to a relative afterthought after ADT variations and Pluvicto within the trial design. Pause and think about that.
Perhaps ironically, the closer our field moves toward treating only visible gross disease with ablative precision - i.e. the closer we get to our own singularity - the more directly we are competing on the same conceptual territory as a radiopharmaceutical delivered through an IV. And the pharmaceutical industry has significant structural advantages in that competition: limited local capital equipment, dramatically lower staffing requirements, scalable manufacturing, and an existing commercial and regulatory infrastructure built for exactly this kind of growth.
Follow the Money
Which brings us to the financial architecture of our field in 2026 - and why the timing of what I’ve described above could not be more consequential.
As we discussed in Part I, finances always play a primary role in shaping the environment. Finances don’t dictate care, but they shape the trajectory of progress, and create either an opportunity or an obstruction - most often a combination of both depending on your perspective.
The 2026 Medicare billing changes have restructured how radiation oncology operates in the United States. With no more than 1 in 3 patients qualifying for the highest complexity daily treatment code - an estimate widely cited from within our own “experts” - the proforma math on new facility investment has fundamentally changed.
Consider buying a new machine at $7 million or more compared to an aging 5-10 year old treatment machine. Your incremental payment is ~$170 within a CMS environment. With 30 patients under beam (clearly high end volume), that results in no more than 10 patients for a gross revenue increase of about $1700. We’ll mix in private pay contracts and bump that to $2500. That’s about $625,000 in additional revenue annually. The treatments are more complex so physician and physics oversight are mandated at a higher level. Your documentation and protocols have incrementally higher costs. Let’s assume (at best) $400,000 flows downstream to the bottom line.
If you are an investor, you’d like to pay off the extra investment in no more than 3 years (ball park general financing number). That puts the MAXIMUM cost difference between the new and older machine around $1.2M - the actual gap is likely closer to $4M. Frankly, it doesn’t work.
Even worse, the opportunity for a new standalone radiation oncology facility, built on the assumption of a technology-forward practice, has largely disappeared within the current framework. I could repeat another simple math example, but again, it lands squarely on a negative outlook.
In earlier decades, this would have mattered less. The timing would have been unfortunate but far more manageable - absorbed into a growing field with expanding indications and rising utilization. But the world has changed over the past 10-20 years, and the financial reality of 2026 is landing at precisely the moment when fuel (money) is pouring into oncology drug development by the hundreds of billions.
The 2002 Medicare IMRT code catalyzed rapid adoption of a technology that transformed our field to a “golden era” of progress - it set the foundation for the massive value we create today. The 2026 changes feel like the mirror image - a structural signal that the economic case for capital-intensive, complexity-driven radiation oncology is under real and present danger.
Money has always carried massive impact. Right now, it is fully aligned with pharmaceutical oncology and offering significant resistance to anything but incremental consolidation in ours. I truly wish that weren’t the case. But it is where we sit.
Quick Aside: The Reflexion platform - a bridge between the old and new.
If you’re not familiar, the Reflexion platform is somewhat of a “bridge” between SBRT and thernostics. It is a PET based system where dose is basically “auto-applied” to uptake. So biology based treatment rather than strictly image based but still externally delivered. To me, it is a rather natural “bridge” in the confluence we are witnessing.
But look at how they went about it.
The interesting part has been their approach to the rollout and their successful acquisition of their new code. I’ve followed them pretty closely for a handful of years - talked to the them along the way - no inside information.
That said, what struck me was their approach for a new radiation machine. Early on, their code requests were almost 100% on the technical side. I remember the phone calls quite well. And from a marketing perspective - where did they go? They went to busy medical oncology practices - to high volume metastatic treating oncologists. That is where they started and they worked backward towards the radiation departments.
And they have captured real income using professional lobbyist and a business focused leadership team. Ultimately there is likely a 10% bump for the radiation oncologist compared to SBRT. But on the technical side reimbursement sits at $3750 per fraction. $3750 per fraction. You (your institution) get $1827 for your tracking and SBRT of a moving lesion (or $560 for your 8 hypofractionation approach if you gate and get lucky - $400 if you are not truly gating and $100 if Evicore deems your metastatic treatment simple). And they just got $3750 per fraction for “biology based” treatment - across the board.
Don’t consider this a negative statement about Reflexion - their story elegantly contrasts the patterns we see elsewhere in our field. They have done what others in our field have failed to accomplish - getting new codes and new appropriate payment rates for new technology that will provide value over pharmaceutical approaches. I personally applaud their results. In a way, if you’re offended at the payment discrepancies above, I think your “offense” is at the lack of appropriate reimbursement for the remainder of what we do.
The Accelerant - AI in labs and research
Further impacting the future is the fact that the trajectory pharmaceutical novel drug development doesn’t move in isolation. It is being actively compressed by AI in ways that would have seemed implausible a decade ago.
Demis Hassabis and the AlphaFold team at Google DeepMind received the Nobel Prize in Chemistry in 2024 for a model that takes a DNA sequence and generates an accurate three-dimensional protein structure - work that previously required years of laboratory effort. The proprietary successor at Isomorphic Labs has pushed the capability significantly further partnering with TRILLION dollar pharmaceutical companies. Their stated mission is, without irony, to “Solve all disease”.
Just so you appreciate the scale - in one hour the system today replaces the work of one PhD student working for 4 years in a top lab. The AI system today is the absolute worst it will ever be. Each month it will become even more efficient making this gap even larger.
The first drug emerging from this style of AI-accelerated pipeline is expected to enter clinical trials later this year. This is not science fiction. This is the current state of the pipeline.
The model compresses years of basic science into silicon calculations. That work will be validated by robotics running 24-hour automated laboratories. At a minimum, the next decade is likely to see 20 to 30 years of prior drug development progress condense into 5 to 7 years. It is very likely that our entire regulatory structure will have to be re-written to address the speed of progress. Hundreds of incremental steps toward controlling and curing cancer - each arriving faster than the last.
Now consider that roughly 80% of survival-improving standard-of-care advances in oncology over the past decade have come from the drug side of the ledger. Four to one, in favor of systemic therapy, before the AI accelerant was applied. Before literally trillions of dollars were added to the other side of the equation. The ratio going forward is almost certain to tilt further.
And just to be clear, this approach will amplify all types of oncology drugs - from the ones to address toxicity, to perhaps new chemotherapies, to novel immunotherapies, as well as to the ligands needed for new theranostics.
This is the context in which radiation oncology’s own singularity is playing out.
Looking towards the horizon:
Our field is moving, inexorably and appropriately, toward treating less uninvolved tissue with more precise ablative doses in fewer fractions. The clinical logic is sound. The data supports it. It is the right direction.
Pharmaceutical theranostics is moving towards the same conceptual target - visible disease, molecular precision, ablative intent - through a fundamentally different and structurally advantaged delivery mechanism.
Further, the upheaval of our financial architecture in 2026 is creating real headwinds against the capital investment required to compete at the leading edge of technology-driven radiation oncology unless you go “outside” and “around” the existing codes that our specialty leans on.
We have three or more trend lines, converging and aligning alongside our own path to Singularity. I think the simple expectation is for the rate of change to significantly accelerate. I think that should be your expectation.
None of this means the field disappears. It means the field transforms - and the question is whether radiation oncologists are active architects of that transformation or late respondents to it. Our history, if we’re honest, suggests we trend toward the latter. Too often, we have been slow to act, even compared to other specialties within a highly regulated clinical medical environment. Too often, we attempt to defend it as “data-driven caution”, but it has been more than that - often compounded by internal finger pointing and side bars questioning our value in the oncology space.
The NEJM review that opened Part I closes with language that is careful and measured in the way academic writing tends to be - but if you read it closely, the message is not subtle:
With modern innovations that reduce the exposure of uninvolved tissues to radiation, the advent of treatment approaches that prioritize highly conformal ablative dosing only to grossly visible disease, and the availability of treatments for side effects of irradiation, radiation oncologists and collaborating physicians have begun to revisit what is defined as an “acceptable risk” of side effects from radiation, with the emerging concept that allowing other wise preventable injury or repairable injury to occur may permit more effective treatment and a more favorable overall side-effect profile. Yet radiation oncologists continue to prioritize avoidance of side effects altogether, a strategy that allows radiotherapy to retain its reputation as an effective and safe treatment for a broad scope of malignant conditions.
If we don’t understand where we sit and begin to take greater ownership over our direction forward - if we don’t move faster and be stronger advocates for our value in oncology - pharma will define our singularity. I don’t believe that is pessimism, I believe it is pattern recognition.
The Ski Lift
A few weeks ago, riding up a lift, I shared a chair with a medical student. We talked about a number of topics before finally getting around to me being in medicine. As the lift was ending, I told her on my view of radiation oncology from my perspective as I peer over the edge towards “retirement”.
It has been a wonderful journey. The impact we have is real and incredibly meaningful. Our ability to help patients is extraordinary. I would choose the same path again without hesitation. It simply is a great spot.
I believe the value we create is not going away. The biology doesn’t change. There are disease sites and clinical scenarios where external beam radiation, delivered with precision and intent, remains and will remain the best available option now and for the foreseeable future. The field does not disappear. But it does change. And the rate of that change is about to accelerate in ways that will feel uncomfortable if we arrive at them unprepared.
Our singularity is not a threat to be feared. It is a direction to be understood - and if we’re paying attention, one that we help in defining. If you are young, at least be cognizant of these trends, and consider how your work aligns or contradicts our fields over-arching direction. Often through history, it is those that prove the current trends wrong who create the most progress, but knowing that you are swimming up stream is critical to that path.
Buckle up, the convergence is underway. It will bring both massive change and opportunity.
As always, thanks for reading. Keep advocating for our value. Keep pushing for better.
The current efforts around radonc billing are like rearranging deck chairs in the titanic. Futility. Sorry but your well written article has made it clear whats up. Or in our case down and eventually out.
The golden era of radonc was circa 2005. We are fine for another 5 to maybe even 10 years, but will decline further. If FLASH proves to be valuable, our specialty will not be salvaged. I'm fortunate to have seen the glorious IMRT and SBRT years but I would harshly steer any newbies to Medonc to avoid the singularity... to be kind. Who knows maybe zietman was right and we get folded into interventional oncology..