There is a tendency to describe advances in nuclear medicine by talking about better images: higher sensitivity, greater resolution, earlier detection and more lesions found. All of those things matter, but I think something more consequential is happening. The value of the image is beginning to move downstream, because it is no longer enough to show physicians something they could not see before. Increasingly, molecular imaging has to tell them something they can act on.
Look at several pieces of research emerging this week. In the five-year PSMAgRT randomized trial, PSMA PET was not simply being evaluated as a better staging tool. Physicians used what they saw on the scan to change radiation treatment, including where radiation was delivered and which areas received additional treatment. The overall trial missed its prespecified primary failure-free survival endpoint, so the findings should not be overstated, but biochemical disease control favored PSMA-guided treatment, particularly in the high-risk and post-prostatectomy populations.
That distinction is important because the question was not whether PSMA PET could detect additional disease. We already have considerable evidence that it can. The more consequential question was whether doing something differently because of the scan could change what happened to the patient, and that is a much higher bar for molecular imaging than simply showing greater diagnostic sensitivity.
At Cleveland Clinic, researchers are taking that concept even closer to the treatment itself. Investigators used PSMA PET to identify localized prostate cancer recurrence and then placed radioactive I-125 brachytherapy seeds into the areas identified by imaging. In the 105-patient study, 79% of patients had no evidence of disease at the most recent follow-up, although the retrospective design and median follow-up of 22.3 months mean those results should still be considered preliminary. What makes the study particularly interesting is not simply the disease-control number, but the role of the scan: the image was helping determine where the radioactive treatment physically went.
We talk constantly about theranostics as the pairing of a diagnostic radiopharmaceutical with a therapeutic radiopharmaceutical directed toward the same biological target. That remains one of the most powerful ideas in nuclear medicine, but I increasingly think the larger transformation extends beyond the traditional diagnostic-therapy pair. The deeper idea is that molecular information can become treatment information.
A PET scan can identify where a target is expressed, but that information can now do much more than confirm the presence of disease. It can determine whether a patient qualifies for therapy, alter a radiation field, identify a lesion for focal treatment or potentially determine which parts of a heterogeneous disease burden deserve different treatment approaches. As those capabilities develop, the image begins to move from an observational role into an operational one.
Researchers are now beginning to attack the other side of the equation as well: seeing the therapy after it has been delivered. Newly published work from Lawrence Berkeley National Laboratory, UC San Francisco and Siemens Healthineers demonstrated first-human imaging of Ac-225 therapy on a clinical time-of-flight PET scanner using cascade gamma rays from the radionuclide’s decay chain. The technique remains experimental and the first clinical demonstration involved only one patient, but the concept matters because Ac-225 has historically presented a difficult post-treatment imaging problem.
Ac-225 is attractive because of what its alpha particles can do over microscopic distances, yet those same characteristics create challenges when physicians want to understand where the therapeutic activity actually went. If that begins to change, the image could eventually become a means of verifying therapy rather than simply identifying disease before treatment. Physicians could begin asking not only where the tumor is, but where the administered activity accumulated, how much reached the tumor, what reached normal organs and whether two patients receiving the same nominal activity experienced meaningfully different biodistribution.
The other part of this transition is happening through software. The newly published LION work trained open-source tumor-segmentation models using more than 7,000 FDG and PSMA PET examinations, with one of the most interesting findings showing that a mixed-disease model trained on only 500 patients could perform comparably to a lymphoma-specific model trained on more than 3,000 cases. That suggests diversity within the training dataset may matter enormously for how well these systems perform outside the environment in which they were developed.
Segmentation can sound like a narrow computer-vision problem, but its importance becomes much clearer when viewed through this broader transition. Once an image becomes reliably computable, clinicians and researchers can move from simply saying that disease is present toward measuring total disease burden, lesion volume, changes over time and potentially treatment-specific response. Combine increasingly sensitive imaging with quantitative analysis, dosimetry and therapeutic radiopharmaceuticals, and nuclear medicine begins generating something richer than pictures: it starts producing structured information that can feed directly into treatment decisions.
That shift also changes the role of imaging data itself. If algorithms are going to influence treatment planning, response assessment or patient selection, then the quality and diversity of the underlying data become part of the clinical infrastructure. The LION study’s finding that a smaller but more varied dataset could rival a much larger disease-specific one suggests that the competitive advantage in nuclear medicine AI may not simply belong to whoever owns the most scans, but to whoever has the data that best represent the clinical world in which those tools must operate.
This evolution has implications across the nuclear medicine economy. For imaging-agent developers, competitive differentiation cannot ultimately stop at sensitivity or lesion detection. The more important question may increasingly become whether a tracer changes management, improves treatment selection or enables an intervention that otherwise would not have occurred.
For scanner companies, the opportunity extends beyond making better pictures. Hardware increasingly becomes the platform through which quantitative disease information, treatment planning and potentially post-therapy assessment are produced. For AI companies, the long-term value is unlikely to come simply from drawing a contour around a lesion faster than a human can; it will come from translating imaging into information clinicians trust enough to use when making treatment decisions.
Radiopharmaceutical developers may face a similar shift. Better visualization of therapeutic biodistribution could eventually feed back into molecule design, dosing strategies, patient selection and treatment sequencing. If developers can understand not only whether a compound binds to the desired target but how therapeutic activity distributes throughout the body in real patients, imaging becomes part of the development loop rather than simply an endpoint along the way.
None of this happens automatically. The PSMAgRT primary endpoint reminds us that finding more disease does not necessarily improve outcomes. The Ac-225 imaging work is a first-human feasibility demonstration, not yet a clinical dosimetry solution, and strong segmentation performance is not the same thing as clinical utility. Those limitations do not weaken the larger trend; they define the evidence that will be needed as nuclear medicine moves closer to the treatment decision.
Nuclear medicine spent decades proving that molecular imaging could reveal biology that conventional imaging could not, and much of that argument has already been won. The next argument is harder: proving that seeing something earlier, more precisely or more quantitatively leads physicians to make a better decision and ultimately produces a better outcome.
That could mean changing a radiation field because PSMA PET revealed disease that otherwise would have been missed, placing brachytherapy seeds into a recurrence defined by molecular imaging or eventually looking at an Ac-225 therapy after administration and determining whether its distribution matches what physicians expected. In each case, the image is no longer simply documenting disease. It is becoming part of the mechanism through which treatment is selected, delivered or evaluated.
That may also change how value is measured across the sector. The commercial winners in this environment may not necessarily be the companies producing the prettiest images or even those detecting the smallest lesions. They may be the technologies that sit closest to a consequential clinical decision, because the future value of nuclear medicine will increasingly depend not only on what the image shows, but on what physicians are able to do because of it.
