Discover how Yttrium-90 radioembolization, AI-assisted dosimetry, multimodal radiotherapy, and patient-specific digital twins are converging to redefine the future of precision radiopharmaceutical medicine.
Disclaimer: This report is intended for informational and educational purposes only and does not constitute medical advice, diagnosis, or treatment. The clinical protocols, AI-driven dosimetry models (e.g., 3D DosiNet), and combined-modality therapies discussed are based on current research and active clinical trials (such as DOORwaY90 and EMERALD-Y90) and may not yet be universally approved or standardized in all jurisdictions. Federal (USA) law restricts the sale and use of $^{90}\text{Y}$ microspheres and associated radiotherapy equipment to qualified physicians, and treatment decisions must be made in consultation with a multidisciplinary oncology team. Markedly abnormal synthetic or excretory liver function remains a contraindication for certain treatments described herein. Readers should consult with a licensed healthcare provider for specific medical guidance regarding liver cancer therapies, safety profiles, and associated risks.
Cancer care is entering a new era in which the boundaries between internal radionuclide therapy, molecular imaging, and external beam radiotherapy are becoming increasingly interconnected. At the center of this transformation is Yttrium-90 (Y-90), a high-energy beta-emitting radionuclide that has become a cornerstone of locoregional liver cancer therapy. Advances in patient-specific dosimetry, molecular imaging, artificial intelligence (AI), and adaptive treatment planning are expanding its role beyond conventional radioembolization toward a more integrated model of precision radiopharmaceutical medicine. Rather than delivering standardized radiation, clinicians are increasingly able to tailor therapy to individual tumor biology, anatomy, and treatment response, moving oncology closer to truly personalized care.
Current Clinical Reality of Y-90 and the Biophysical LandscapeYttrium-90 transarterial radioembolization (TARE), also known as selective internal radiation therapy (SIRT), has evolved from a salvage treatment into an established locoregional therapy for carefully selected patients with unresectable hepatocellular carcinoma (HCC) and liver-dominant metastatic colorectal cancer. Its clinical effectiveness is rooted in its unique physical properties. As a pure beta emitter, Y-90 delivers high-energy radiation with a maximum beta energy of approximately 2.27 MeV and a physical half-life of about 64 hours. The emitted particles have an average tissue penetration of roughly 2.5 mm and a maximum range of approximately 11 mm, enabling highly localized irradiation while minimizing exposure to surrounding healthy tissue.
These characteristics make Y-90 particularly well suited for liver-directed therapy. Administered through the hepatic artery, Y-90 microspheres preferentially accumulate within hypervascular tumors, which derive most of their blood supply from the hepatic arterial circulation, while relatively sparing normal liver parenchyma supplied primarily by the portal vein. This selective vascular distribution underpins the therapeutic advantage of radioembolization and has established Y-90 as an important component of modern multidisciplinary liver cancer management.
Microsphere Technologies: Glass Versus Resin
Clinical Y-90 therapy is delivered using two commercially available microsphere platforms: TheraSphere® (glass microspheres) and SIR-Spheres® (resin microspheres). Although both employ the same radionuclide, important differences in material composition, specific activity, particle number, and embolic characteristics influence treatment planning and clinical application.
Glass microspheres carry substantially higher specific activity per sphere, allowing the prescribed radiation dose to be delivered with relatively few particles and generally producing minimal embolic effect. Resin microspheres contain considerably lower activity per sphere, requiring many more particles to achieve the intended dose. The larger particle burden may increase embolic effects during treatment, which can influence microsphere distribution depending on individual tumor vascularity and treatment objectives.
An important milestone for resin microspheres was the DOORwaY90 clinical trial, which demonstrated encouraging local tumor control and contributed to the expanded regulatory indication for SIR-Spheres in unresectable hepatocellular carcinoma. In July 2025, the U.S. Food and Drug Administration approved SIR-Spheres for local tumor control in appropriately selected patients with unresectable HCC, representing an important expansion of treatment options for liver-directed therapy.
At a Glance: Key CharacteristicsResin Microspheres (SIR-Spheres®)
Glass Microspheres (TheraSphere®)
Guideline Evolution and Patient Selection
The expanding role of Y-90 radioembolization within contemporary treatment guidelines—including recommendations from the National Comprehensive Cancer Network (NCCN) and the Barcelona Clinic Liver Cancer (BCLC) framework—reflects growing evidence supporting personalized treatment strategies. Recent guideline updates increasingly emphasize individualized dosimetry, preservation of functional liver reserve, and multidisciplinary patient selection rather than standardized activity administration.
Appropriate patient selection remains fundamental to safe and effective treatment. Clinical evaluation typically incorporates liver function, performance status, tumor burden, vascular anatomy, and lung radiation exposure. Common selection parameters include preserved liver function, acceptable bilirubin levels, Child–Pugh class A or carefully selected B7 patients, and favorable hepatic vascular anatomy. Assessment of lung shunt fraction and estimated lung absorbed dose remains essential because excessive hepatopulmonary shunting can increase the risk of radiation pneumonitis. Rather than relying on fixed thresholds alone, contemporary practice increasingly integrates patient-specific dosimetry with multidisciplinary clinical judgment to optimize both treatment efficacy and safety.
The Role of AI in Transforming Radiotheranostics
Artificial intelligence is emerging as an important enabler of precision radiotherapy and radiopharmaceutical medicine. By helping clinicians interpret complex imaging data, estimate absorbed-dose distributions, and reduce the computational demands of advanced dosimetry, AI may strengthen decision-making across the treatment pathway—from pre-treatment planning and activity prediction to procedural guidance and post-treatment assessment.
Its most immediate value lies not in replacing established physics-based methods or clinical judgment, but in making complex calculations faster, more reproducible, and more responsive to patient-specific anatomy and tumor behavior.
Generative AI and Dose Prediction
Generative models, including Generative Adversarial Networks (GANs), are being investigated for three-dimensional dose prediction, image synthesis, and activity-map generation. In Y-90 radioembolization, experimental models can use pre-treatment technetium-99m macroaggregated albumin (Tc-99m MAA) imaging to estimate the probable post-treatment distribution of Y-90 microspheres.
Because Tc-99m MAA is administered during treatment planning as a surrogate for microsphere deposition, AI-based analysis may help identify patterns that conventional interpretation does not fully capture. These models could support improved estimates of tumor uptake, normal-liver exposure, and spatial dose heterogeneity before therapy is delivered.
Voxel-based deep-learning approaches, including architectures derived from Pix2Pix and related image-to-image translation frameworks, have shown encouraging performance in research settings. Some studies have reported close agreement between predicted and reference dose distributions, although results depend heavily on imaging quality, registration accuracy, cohort size, and the method used to generate the reference data.
Generative systems are also being explored for synthetic CT, PET, and SPECT image generation. These images may help compensate for missing modalities, improve attenuation correction, or represent tissue heterogeneity more effectively. However, synthetic imaging remains an investigational application and requires rigorous external validation before it can be relied upon for clinical treatment decisions.
Deep Reinforcement Learning and Treatment Optimization
While generative models are primarily designed to predict images or dose distributions, deep reinforcement learning (DRL) is being investigated as a method for optimizing treatment decisions.
In reinforcement-learning systems, an algorithm evaluates repeated treatment simulations and learns which actions are most likely to achieve a defined objective. In radiotherapy, that objective may involve maximizing tumor coverage while limiting radiation exposure to healthy tissue and organs at risk.
This approach may be particularly relevant when Y-90 radioembolization is combined with external-beam techniques such as stereotactic body radiotherapy (SBRT). Internal radionuclide therapy can produce highly heterogeneous dose distributions because microsphere deposition depends on vascular anatomy and blood flow. External radiation may then be planned to compensate for regions receiving insufficient internal dose.
Interactive planning systems represent another emerging direction. In these systems, clinicians can adjust the relative importance of tumor coverage, liver preservation, and organ-at-risk constraints through intuitive controls. The underlying optimization engine then generates a revised treatment plan reflecting those preferences.
Although such systems may eventually support faster and more individualized planning, most remain at the research or prototype stage. Their clinical value will depend on transparent objective functions, robust validation, and the preservation of meaningful physician oversight.
Computational Dosimetry: From Monte Carlo to GPU Acceleration
Accurate dosimetry is one of the most computationally demanding components of radiopharmaceutical therapy. Monte Carlo simulation is widely regarded as a reference method because it models the physical transport and interaction of radiation within heterogeneous tissues. Its principal limitation has traditionally been computational cost, with detailed simulations often requiring substantial processing time.
GPU-accelerated computing is beginning to reduce this barrier. By performing many calculations in parallel, graphics-processing units can substantially shorten Monte Carlo runtimes while preserving much of the method’s physical accuracy. This creates the possibility of incorporating advanced dose calculations into more practical clinical workflows.
Deep-learning dose engines offer an additional approach. Models based on three-dimensional convolutional networks, residual networks, and U-Net architectures can be trained to approximate dose maps generated by Monte Carlo simulation or other physics-based methods. Systems sometimes described as 3D DosiNet models aim to reproduce patient-specific absorbed-dose distributions substantially faster than conventional simulation.
The performance of these models should be evaluated using clearly defined metrics, such as voxel-level dose error, dose-volume histogram agreement, organ-level absorbed-dose differences, and spatial gamma analysis. Processing speed alone is not sufficient; the model must also demonstrate reliability across different scanners, institutions, patient anatomies, acquisition protocols, and disease patterns.
A Flexible Dosimetry Ecosystem
Contemporary radiopharmaceutical dosimetry includes several complementary approaches:
MIRD-based dosimetry provides relatively rapid organ- or compartment-level absorbed-dose estimates. It remains useful when computational resources or detailed voxel-level imaging are limited.
Voxel-based S-value methods estimate absorbed dose at a finer spatial scale and can better represent heterogeneous activity distributions, although their accuracy depends on image resolution and kernel assumptions.
Conventional Monte Carlo simulation provides highly detailed modeling of radiation transport and tissue interaction but may require substantial computational time and expertise.
GPU-accelerated Monte Carlo methods seek to preserve the physical rigor of Monte Carlo simulation while reducing processing time sufficiently for more practical clinical use.
Deep-learning dose models can generate rapid approximations of absorbed-dose distributions, but their reliability depends on the quality and diversity of the training data and the strength of external validation.
These approaches should not be viewed as mutually exclusive. A future clinical workflow may use rapid AI-generated estimates for initial planning, followed by physics-based verification in higher-risk or technically complex cases.
Clinical Significance and Limitations
As these technologies mature, they may help reduce uncertainty in estimates of tumor dose, healthy-liver exposure, and lung radiation burden. They may also make patient-specific dosimetry more accessible in institutions where computational resources or specialist expertise are limited.
However, AI-generated predictions remain sensitive to imaging artifacts, segmentation errors, scanner differences, incomplete training data, and changes in clinical practice. Models trained at one institution may not perform consistently in another without recalibration or external validation.
The broader significance is therefore not that AI will replace radiotheranostics, medical physics, or multidisciplinary clinical expertise. Its value lies in helping these disciplines operate with greater speed, consistency, and patient specificity. Used responsibly, AI can serve as a computational layer that strengthens precision treatment while preserving transparent validation, quality assurance, and human oversight.
Clinical Synergy: Integrating Y-90 with External-Beam and Molecular Imaging Modalities
One of the most promising directions in contemporary radiotheranostics is the integration of internal Yttrium-90 (Y-90) radioembolization with external-beam radiation techniques such as stereotactic body radiotherapy (SBRT) and external-beam radiation therapy (EBRT).
Y-90 microsphere distribution is governed by hepatic arterial anatomy, blood flow, catheter position, and tumor perfusion. As a result, the absorbed-dose distribution can be highly heterogeneous. Some regions may receive a substantial tumoricidal dose, while poorly perfused areas receive less radiation. These relatively underdosed regions are sometimes described as cold spots.
External-beam radiotherapy offers a complementary mechanism. Because it does not depend on intra-arterial microsphere deposition, EBRT or SBRT can be directed toward residual disease or regions that received insufficient internal radiation. In principle, this creates a combined strategy in which Y-90 delivers concentrated intra-arterial treatment and external radiation supplies a more spatially controlled dose to selected targets.
The clinical challenge is not simply to add the two treatments together. Their biological effects, spatial dose distributions, timing, treated liver volumes, and cumulative exposure to healthy tissue must be evaluated carefully. Effective integration therefore requires multimodality image registration, patient-specific dosimetry, and multidisciplinary treatment planning.
The Safety of Sequential Radiation
One of the principal concerns surrounding sequential Y-90 and external-beam treatment is cumulative liver toxicity. Patients undergoing radioembolization may already have cirrhosis, compromised functional liver reserve, previous systemic treatment, or extensive tumor burden. Additional radiation must therefore be planned around the amount and condition of the remaining healthy liver.
A 2026 retrospective study led by University of Cincinnati Cancer Center investigators reviewed 94 patients who received liver-directed EBRT, including 15 who had previously undergone Y-90 treatment. The investigators did not observe an increase in liver toxicity among the patients receiving EBRT after Y-90 and concluded that carefully individualized external radiation remained feasible in this selected population. The study provides meaningful reassurance, although its retrospective design and relatively small Y-90 subgroup mean that it should not be interpreted as establishing universal safety.
These findings support further investigation of individualized treatment sequences. Y-90 may be used first to achieve local control, reduce viable tumor volume, facilitate downstaging, or support transplantation strategies. Focused SBRT or EBRT may subsequently be considered for residual, recurrent, or insufficiently treated disease.
The decision must remain patient-specific. Important variables include baseline liver function, prior absorbed dose, treated liver volume, time between therapies, vascular anatomy, tumor location, and the dose delivered to uninvolved liver and adjacent organs.
Theranostic Pairs and Molecular Precision
Theranostics combines diagnostic imaging and targeted therapy through radiopharmaceuticals that share the same biological targeting mechanism. A diagnostic radionuclide is used to visualize and quantify target expression or biodistribution, while a therapeutic radionuclide delivers radiation to the same molecular target.
Some theranostic systems use different isotopes of the same element. Others use different radionuclides attached to the same or closely related targeting ligand. The objective is to determine whether a tumor expresses the intended target, estimate where the therapeutic compound is likely to accumulate, and support individualized activity selection and dosimetry.
The Y-86/Y-90 pair illustrates the matched-isotope concept. Y-86 is a positron-emitting radionuclide that can be imaged with PET, while Y-90 delivers therapeutic beta radiation. Preclinical research has demonstrated the use of Y-86 imaging to evaluate the biodistribution of Y-90-labeled targeting compounds. However, this approach remains distinct from routine Y-90 microsphere planning and should not be presented as an established component of liver radioembolization practice.
Radiotheranostic research is also expanding beyond established targets such as somatostatin receptors and prostate-specific membrane antigen. Fibroblast activation protein, or FAP, is being investigated because it is expressed by cancer-associated fibroblasts in the microenvironment of many solid tumors.
A 2024 preclinical study evaluated a Ga-68/Lu-177-labeled FAP-targeting compound known as FAPT. Ga-68 supported PET imaging, while Lu-177 provided therapeutic beta emission. The results demonstrated increased tumor uptake and prolonged retention in experimental models, but the compound remains investigational and should not yet be described as a broadly available clinical theranostic platform.
These developments illustrate the broader logic of molecular radiotheranostics: imaging can help determine whether a therapeutic target is present, quantify its distribution, and provide evidence for selecting or excluding a particular treatment.
Synergy with Immunotherapy
The interaction between radiation and the immune system is another important area of investigation. Radiation can cause tumor-cell injury, alter inflammatory signaling, release tumor-associated antigens, and reshape the local tumor microenvironment. These effects create a biological rationale for combining radiotherapy with immune checkpoint inhibition.
Both Y-90 radioembolization and SBRT can produce localized radiation effects within liver tumors. Researchers are investigating whether these effects can increase immune recognition or improve the activity of systemic immunotherapy in selected patients.
The EMERALD-Y90 study is evaluating transarterial radioembolization in combination with durvalumab and bevacizumab for hepatocellular carcinoma. The trial is designed to examine the safety and clinical activity of combining liver-directed radiation with immune checkpoint inhibition and antiangiogenic therapy. It should be described as an active clinical investigation rather than an established treatment pathway.
At present, it remains premature to conclude that Y-90 combined with immunotherapy improves overall survival. Clinical benefit must be demonstrated through appropriately designed prospective trials that distinguish the contribution of radioembolization from the effects of systemic therapy.
The most important questions include treatment sequencing, patient selection, radiation dose, liver function, immune-related toxicity, and whether local radiation can produce clinically meaningful systemic immune effects.
The Regulatory and Standardization Frontier
As artificial intelligence, advanced dosimetry, and new radiopharmaceutical combinations move toward clinical use, validation and standardization become increasingly important. A model that performs well in one retrospective dataset may not retain the same accuracy across different scanners, institutions, acquisition protocols, patient populations, or clinical workflows.
Innovation alone is therefore insufficient. AI-supported radiotherapy systems must also be reproducible, explainable within their intended context, continuously monitored, and subjected to appropriate quality assurance.
ESTRO–AAPM Guidelines for AIIn 2024, the European Society for Radiotherapy and Oncology and the American Association of Physicists in Medicine published a joint guideline addressing the development, clinical validation, and reporting of AI models in radiation therapy.
Developed through a Delphi consensus process, the guideline contains 19 statements covering major requirements for responsible AI development and clinical translation.
Its central priorities include:
Transparent reporting
The intended clinical use, model architecture, data sources, training procedures, validation methods, performance metrics, and known limitations should be described clearly.
Representative data
Development and validation datasets should reflect the clinical populations, imaging systems, and treatment conditions in which the model is intended to operate.
Independent validation
Performance should be evaluated outside the original development environment whenever possible. Internal validation alone cannot establish broad generalizability.
Clinical relevance
Technical accuracy must be connected to clinically meaningful outcomes. A statistically strong prediction is not automatically useful if it does not improve treatment decisions, workflow, safety, or patient outcomes.
Quality assurance and monitoring
AI systems require testing before deployment and continued surveillance after implementation. Performance can change as imaging equipment, clinical protocols, patient populations, or software environments evolve.
Together, these principles create a bridge between experimental model development and dependable clinical adoption.
Regulatory Engagement and FDA Guidance
In January 2025, the U.S. Food and Drug Administration issued draft guidance on the use of AI-generated information to support regulatory decision-making for drugs and biological products.
The guidance proposes a risk-based credibility framework tied to a model’s specific context of use. It asks developers to define the question being addressed, establish the intended role of the model, assess the risks associated with incorrect outputs, document credibility evidence, and determine whether model performance is sufficient for that particular application.
The scope of this guidance must be interpreted carefully. It primarily addresses AI used to produce information supporting regulatory decisions about drugs and biological products. It is not, by itself, a complete regulatory framework for every AI-enabled dosimetry or treatment-planning system.
Software used to calculate, recommend, or control patient treatment may be regulated as a medical device or as Software as a Medical Device, depending on its intended use and functionality. Such systems may be subject to separate device-specific requirements governing verification, validation, cybersecurity, software maintenance, human factors, and post-market monitoring.
The broader regulatory principle is nevertheless clear: the level of evidence should correspond to the clinical risk. A model used for research prioritization does not require the same safeguards as an algorithm whose output directly influences the radiation activity administered to a patient.
Infrastructure, Global Access, and Low- and Middle-Income Countries
The development of precision radiotherapy cannot be separated from the problem of unequal access. Advanced dosimetry and AI-supported planning have limited practical value in regions that lack basic imaging systems, radiotherapy equipment, trained professionals, reliable isotope supply, or sustainable maintenance programs.
The access gap remains substantial. The IAEA has reported that while radiotherapy is broadly available in many high-income countries, access is significantly lower in middle-income countries and may reach only a small proportion of patients in low-income settings.
Closing this gap requires more than donating equipment. Sustainable radiotherapy systems depend on trained clinicians, medical physicists, radiation therapists, radiochemists, engineers, regulatory capacity, quality-assurance programs, maintenance infrastructure, and long-term financial planning.
The IAEA Rays of Hope Initiative
The International Atomic Energy Agency launched the Rays of Hope initiative in February 2022 to support countries establishing or expanding medical imaging, nuclear medicine, and radiotherapy services. The program prioritizes high-impact interventions designed around each participating country’s needs and existing healthcare capacity.
Its work includes:
A major milestone occurred in July 2025, when Malawi opened its first public radiotherapy center at Kamuzu Central Hospital in Lilongwe. The center was supported through the Rays of Hope initiative and expanded domestic access to treatment that had previously been extremely limited.
The Malawi center demonstrates that infrastructure development can produce measurable improvements in access. It also illustrates why equipment must be accompanied by workforce training, data collection, safety systems, and sustained technical support.
Decentralized Isotope Production and the Role of Accelerators
Medical isotope supply is another major component of global access. Many radionuclides depend on specialized reactors, generators, cyclotrons, linear accelerators, target-processing facilities, and international distribution networks. Short physical half-lives and complex radiochemistry make some products especially difficult to transport over long distances.
Accelerator-based production may help diversify supply for selected radionuclides. Medical cyclotrons are already essential for producing short-lived PET isotopes such as fluorine-18 and can support regional diagnostic imaging networks. Compact accelerator systems are also being investigated for the production of additional diagnostic and therapeutic radionuclides.
However, isotope production is highly radionuclide-specific. Cyclotrons and compact linear accelerators cannot be treated as universal replacements for nuclear reactors.
Production feasibility depends on:
Y-90 deserves particular clarification. Clinical Y-90 is commonly obtained through the decay of its parent radionuclide, strontium-90, using generator-based production and specialized processing. A conventional hospital cyclotron or compact radiotherapy linac cannot simply replace this supply chain.
Accelerators can nevertheless strengthen the broader radiotheranostic ecosystem by supporting local or regional production of suitable PET, SPECT, alpha-emitting, or beta-emitting radionuclides where technically and economically feasible.
Production Infrastructure in Context
Nuclear reactors
Reactors remain important for producing many neutron-rich medical radionuclides and parent isotopes. They can provide high yields but require substantial national or regional infrastructure, regulatory oversight, waste management, and long-term maintenance.
Medical cyclotrons
Cyclotrons are well suited to producing many proton-rich diagnostic radionuclides, particularly short-lived PET isotopes. They can be installed at large hospitals or regional production centers but still require target systems, radiochemistry laboratories, quality-control capabilities, shielding, and trained personnel.
Linear accelerators
Specialized proton, ion, or electron linacs can produce selected medical radionuclides through particle- or photon-induced reactions. Their suitability depends on beam characteristics and the isotope being produced. Although they may offer advantages for certain production pathways, claims of universally lower cost, lower waste, or easier maintenance should be assessed on an isotope-specific basis.
A resilient global isotope strategy will likely use a distributed combination of reactors, generators, cyclotrons, linacs, processing centers, and regional distribution networks rather than relying on a single production technology.
Canada’s Contribution
Canada announced its first in-kind contributions to the IAEA’s global Rays of Hope activities in 2026 through work coordinated with the Canadian Nuclear Isotope Council.
The planned contribution includes an isotope-production and radiation-safety training program primarily hosted at Bruce Power. It is intended to provide regulators and specialists from low- and middle-income countries with practical exposure to isotope production, processing, and radiation-safety systems. Additional proposed clinical training includes radiochemistry, radiolabeling, quality control, dosimetry, medical imaging, and patient delivery.
The significance of this initiative extends beyond isotope supply. Long-term self-sufficiency requires regulatory knowledge, clinical expertise, production competence, safety culture, and the ability to maintain quality across the entire pathway from radionuclide generation to patient treatment.
Precision Radiopharmaceutical Medicine: The 2030 Vision
Looking toward the end of the decade, the convergence of radionuclide therapy, molecular imaging, external-beam radiation, AI-assisted analysis, and longitudinal clinical data points toward a more adaptive model of cancer treatment.
In this model, radiotherapy would no longer be planned as a series of isolated interventions. Instead, imaging, absorbed-dose measurements, laboratory biomarkers, treatment response, toxicity, and disease progression would be incorporated into a continuously updated decision process.
The goal would not be to automate oncology. It would be to give multidisciplinary teams a more complete and dynamic representation of the patient’s disease and the consequences of each treatment decision.
The Rise of the Digital Twin
A digital twin is more than a static three-dimensional reconstruction or a collection of patient data. In its strongest form, it is a patient-specific computational model that is repeatedly updated as new observations become available.
An oncology digital twin could potentially combine:
The system could then simulate possible treatment scenarios and compare their expected benefits and risks.
For example, a strong response to Y-90 radioembolization might support observation, systemic therapy, transplantation assessment, or a reduced external-beam treatment volume. Persistent viable disease or an underdosed region might support a focused SBRT boost, repeat locoregional treatment, or an alternative therapy.
This concept remains aspirational. Most current systems described as digital twins are more accurately characterized as multimodal predictive models or longitudinal digital representations. A clinically meaningful digital twin would require repeated calibration, prospective validation, uncertainty quantification, and evidence that twin-informed decisions improve outcomes.
Until those requirements are met, the more precise term is digital-twin-inspired treatment planning.
Integrating Biology at a Finer Scale
Future treatment planning may also incorporate biological features that are not visible through anatomy alone. Tumors frequently contain regions with different levels of perfusion, hypoxia, cellular density, metabolism, immune activity, and treatment resistance.
Multiparametric imaging and AI-supported analysis may help identify these subregions. Instead of prescribing a uniform dose to the entire visible tumor, clinicians could potentially increase radiation to resistant areas while reducing exposure to sensitive normal tissue.
This strategy is often described as biological dose painting.
For biological dose painting to become clinically dependable, imaging biomarkers must be reproducible and linked to meaningful biological characteristics. Treatment-planning systems must also account for spatial uncertainty, motion, changes between imaging and treatment, and the possibility that tumor biology evolves over time.
The objective is not simply greater technical resolution. It is to connect measurable biological differences with treatment decisions that produce better outcomes.
Toward Long-Term Disease Control
The broader objective of adaptive radiotheranostics is not to maximize radiation at every opportunity. It is to achieve durable disease control while preserving liver function, limiting toxicity, and maintaining quality of life.
This may involve different treatment combinations for different patients:
The future of precision radiation medicine will therefore depend less on a single dominant technology and more on the ability to coordinate multiple therapies around the evolving condition of the patient.
Synthesis and Strategic Recommendations
The next generation of radiopharmaceutical medicine will depend on the convergence of radiation physics, molecular targeting, computational modeling, clinical expertise, and sustainable healthcare infrastructure.
Several priorities stand out.
Expand Multi-Institutional Data Sharing
AI models require large, diverse, and well-characterized datasets. Multi-institutional collaboration is essential for determining whether a model performs reliably across different populations, scanners, microsphere platforms, imaging protocols, and treatment practices.
Data-sharing initiatives should include standardized definitions, imaging metadata, segmentation protocols, absorbed-dose calculations, treatment parameters, clinical outcomes, and toxicity reporting.
Privacy-preserving approaches, federated learning, and common data models may help institutions collaborate without transferring all patient-level data into a single centralized repository.
Validate Combined-Modality Treatment Prospectively
Retrospective evidence suggests that selected patients can receive external-beam radiation after Y-90 without unacceptable toxicity, but larger prospective studies are needed.
Future trials should define:
Randomized or carefully controlled studies will be especially important when evaluating combinations involving immunotherapy.
Require Transparent AI Validation
AI dosimetry systems should be evaluated against appropriate physics-based reference methods and tested outside their original development institutions.
Validation should include technical accuracy, clinical relevance, failure detection, subgroup performance, workflow integration, and post-deployment monitoring.
Models should communicate uncertainty rather than presenting every prediction as equally reliable.
Build Isotope-Specific Supply Strategies
Infrastructure planning should be based on the radionuclides a region intends to produce and use. Reactor, generator, cyclotron, and accelerator systems should not be compared only through broad claims about cost or safety.
Each production route requires analysis of yield, purity, target supply, radiochemistry, waste, workforce, distribution distance, and clinical demand.
Regional production hubs may provide a more realistic model than attempting to place every production technology within every hospital.
Strengthen Global Training Standards
As radiotherapy becomes more computationally intensive, education must extend beyond isotope administration and equipment operation.
Clinicians, physicists, radiochemists, technologists, regulators, and engineers will need training in:
Frameworks such as the ESTRO–AAPM guideline and the IAEA Rays of Hope Anchor Centres provide foundations for building this expertise internationally.
ConclusionBy the early 2030s, the most advanced radiotherapy pathways may combine molecular targeting, patient-specific dosimetry, external-beam precision, longitudinal imaging, and computational decision support within one adaptive model of care.
The deeper goal is not technological complexity for its own sake. It is to deliver the appropriate radiation to the appropriate target at the appropriate time while preserving healthy tissue, maintaining clinical accountability, and extending access across very different healthcare systems.
The future of radiotheranostics will be defined not only by what can be calculated or delivered, but by whether those capabilities can be validated, governed, and made meaningfully available to patients.
Recommended Reading
Combined Radiation and Immunotherapy
Theranostic Pairs
AI Validation and Regulation
Global Access and Infrastructure
Current Clinical Reality of Y-90 and the Biophysical LandscapeYttrium-90 transarterial radioembolization (TARE), also known as selective internal radiation therapy (SIRT), has evolved from a salvage treatment into an established locoregional therapy for carefully selected patients with unresectable hepatocellular carcinoma (HCC) and liver-dominant metastatic colorectal cancer. Its clinical effectiveness is rooted in its unique physical properties. As a pure beta emitter, Y-90 delivers high-energy radiation with a maximum beta energy of approximately 2.27 MeV and a physical half-life of about 64 hours. The emitted particles have an average tissue penetration of roughly 2.5 mm and a maximum range of approximately 11 mm, enabling highly localized irradiation while minimizing exposure to surrounding healthy tissue.
These characteristics make Y-90 particularly well suited for liver-directed therapy. Administered through the hepatic artery, Y-90 microspheres preferentially accumulate within hypervascular tumors, which derive most of their blood supply from the hepatic arterial circulation, while relatively sparing normal liver parenchyma supplied primarily by the portal vein. This selective vascular distribution underpins the therapeutic advantage of radioembolization and has established Y-90 as an important component of modern multidisciplinary liver cancer management.
Microsphere Technologies: Glass Versus Resin
Clinical Y-90 therapy is delivered using two commercially available microsphere platforms: TheraSphere® (glass microspheres) and SIR-Spheres® (resin microspheres). Although both employ the same radionuclide, important differences in material composition, specific activity, particle number, and embolic characteristics influence treatment planning and clinical application.
Glass microspheres carry substantially higher specific activity per sphere, allowing the prescribed radiation dose to be delivered with relatively few particles and generally producing minimal embolic effect. Resin microspheres contain considerably lower activity per sphere, requiring many more particles to achieve the intended dose. The larger particle burden may increase embolic effects during treatment, which can influence microsphere distribution depending on individual tumor vascularity and treatment objectives.
An important milestone for resin microspheres was the DOORwaY90 clinical trial, which demonstrated encouraging local tumor control and contributed to the expanded regulatory indication for SIR-Spheres in unresectable hepatocellular carcinoma. In July 2025, the U.S. Food and Drug Administration approved SIR-Spheres for local tumor control in appropriately selected patients with unresectable HCC, representing an important expansion of treatment options for liver-directed therapy.
At a Glance: Key CharacteristicsResin Microspheres (SIR-Spheres®)
- Biocompatible resin microspheres
- Lower specific activity per microsphere (approximately 50 Bq at calibration)
- Diameter approximately 20–60 μm
- Higher particle count with greater embolic potential
- FDA-approved for selected patients with unresectable hepatocellular carcinoma and metastatic colorectal liver metastases
Glass Microspheres (TheraSphere®)
- Glass microspheres
- Higher specific activity per microsphere (approximately 2,500 Bq at calibration)
- Diameter approximately 20–30 μm
- Lower particle count with minimal embolic effect
- FDA-approved for unresectable hepatocellular carcinoma, with additional use in selected cholangiocarcinoma patients
Guideline Evolution and Patient Selection
The expanding role of Y-90 radioembolization within contemporary treatment guidelines—including recommendations from the National Comprehensive Cancer Network (NCCN) and the Barcelona Clinic Liver Cancer (BCLC) framework—reflects growing evidence supporting personalized treatment strategies. Recent guideline updates increasingly emphasize individualized dosimetry, preservation of functional liver reserve, and multidisciplinary patient selection rather than standardized activity administration.
Appropriate patient selection remains fundamental to safe and effective treatment. Clinical evaluation typically incorporates liver function, performance status, tumor burden, vascular anatomy, and lung radiation exposure. Common selection parameters include preserved liver function, acceptable bilirubin levels, Child–Pugh class A or carefully selected B7 patients, and favorable hepatic vascular anatomy. Assessment of lung shunt fraction and estimated lung absorbed dose remains essential because excessive hepatopulmonary shunting can increase the risk of radiation pneumonitis. Rather than relying on fixed thresholds alone, contemporary practice increasingly integrates patient-specific dosimetry with multidisciplinary clinical judgment to optimize both treatment efficacy and safety.
The Role of AI in Transforming Radiotheranostics
Artificial intelligence is emerging as an important enabler of precision radiotherapy and radiopharmaceutical medicine. By helping clinicians interpret complex imaging data, estimate absorbed-dose distributions, and reduce the computational demands of advanced dosimetry, AI may strengthen decision-making across the treatment pathway—from pre-treatment planning and activity prediction to procedural guidance and post-treatment assessment.
Its most immediate value lies not in replacing established physics-based methods or clinical judgment, but in making complex calculations faster, more reproducible, and more responsive to patient-specific anatomy and tumor behavior.
Generative AI and Dose Prediction
Generative models, including Generative Adversarial Networks (GANs), are being investigated for three-dimensional dose prediction, image synthesis, and activity-map generation. In Y-90 radioembolization, experimental models can use pre-treatment technetium-99m macroaggregated albumin (Tc-99m MAA) imaging to estimate the probable post-treatment distribution of Y-90 microspheres.
Because Tc-99m MAA is administered during treatment planning as a surrogate for microsphere deposition, AI-based analysis may help identify patterns that conventional interpretation does not fully capture. These models could support improved estimates of tumor uptake, normal-liver exposure, and spatial dose heterogeneity before therapy is delivered.
Voxel-based deep-learning approaches, including architectures derived from Pix2Pix and related image-to-image translation frameworks, have shown encouraging performance in research settings. Some studies have reported close agreement between predicted and reference dose distributions, although results depend heavily on imaging quality, registration accuracy, cohort size, and the method used to generate the reference data.
Generative systems are also being explored for synthetic CT, PET, and SPECT image generation. These images may help compensate for missing modalities, improve attenuation correction, or represent tissue heterogeneity more effectively. However, synthetic imaging remains an investigational application and requires rigorous external validation before it can be relied upon for clinical treatment decisions.
Deep Reinforcement Learning and Treatment Optimization
While generative models are primarily designed to predict images or dose distributions, deep reinforcement learning (DRL) is being investigated as a method for optimizing treatment decisions.
In reinforcement-learning systems, an algorithm evaluates repeated treatment simulations and learns which actions are most likely to achieve a defined objective. In radiotherapy, that objective may involve maximizing tumor coverage while limiting radiation exposure to healthy tissue and organs at risk.
This approach may be particularly relevant when Y-90 radioembolization is combined with external-beam techniques such as stereotactic body radiotherapy (SBRT). Internal radionuclide therapy can produce highly heterogeneous dose distributions because microsphere deposition depends on vascular anatomy and blood flow. External radiation may then be planned to compensate for regions receiving insufficient internal dose.
Interactive planning systems represent another emerging direction. In these systems, clinicians can adjust the relative importance of tumor coverage, liver preservation, and organ-at-risk constraints through intuitive controls. The underlying optimization engine then generates a revised treatment plan reflecting those preferences.
Although such systems may eventually support faster and more individualized planning, most remain at the research or prototype stage. Their clinical value will depend on transparent objective functions, robust validation, and the preservation of meaningful physician oversight.
Computational Dosimetry: From Monte Carlo to GPU Acceleration
Accurate dosimetry is one of the most computationally demanding components of radiopharmaceutical therapy. Monte Carlo simulation is widely regarded as a reference method because it models the physical transport and interaction of radiation within heterogeneous tissues. Its principal limitation has traditionally been computational cost, with detailed simulations often requiring substantial processing time.
GPU-accelerated computing is beginning to reduce this barrier. By performing many calculations in parallel, graphics-processing units can substantially shorten Monte Carlo runtimes while preserving much of the method’s physical accuracy. This creates the possibility of incorporating advanced dose calculations into more practical clinical workflows.
Deep-learning dose engines offer an additional approach. Models based on three-dimensional convolutional networks, residual networks, and U-Net architectures can be trained to approximate dose maps generated by Monte Carlo simulation or other physics-based methods. Systems sometimes described as 3D DosiNet models aim to reproduce patient-specific absorbed-dose distributions substantially faster than conventional simulation.
The performance of these models should be evaluated using clearly defined metrics, such as voxel-level dose error, dose-volume histogram agreement, organ-level absorbed-dose differences, and spatial gamma analysis. Processing speed alone is not sufficient; the model must also demonstrate reliability across different scanners, institutions, patient anatomies, acquisition protocols, and disease patterns.
A Flexible Dosimetry Ecosystem
Contemporary radiopharmaceutical dosimetry includes several complementary approaches:
MIRD-based dosimetry provides relatively rapid organ- or compartment-level absorbed-dose estimates. It remains useful when computational resources or detailed voxel-level imaging are limited.
Voxel-based S-value methods estimate absorbed dose at a finer spatial scale and can better represent heterogeneous activity distributions, although their accuracy depends on image resolution and kernel assumptions.
Conventional Monte Carlo simulation provides highly detailed modeling of radiation transport and tissue interaction but may require substantial computational time and expertise.
GPU-accelerated Monte Carlo methods seek to preserve the physical rigor of Monte Carlo simulation while reducing processing time sufficiently for more practical clinical use.
Deep-learning dose models can generate rapid approximations of absorbed-dose distributions, but their reliability depends on the quality and diversity of the training data and the strength of external validation.
These approaches should not be viewed as mutually exclusive. A future clinical workflow may use rapid AI-generated estimates for initial planning, followed by physics-based verification in higher-risk or technically complex cases.
Clinical Significance and Limitations
As these technologies mature, they may help reduce uncertainty in estimates of tumor dose, healthy-liver exposure, and lung radiation burden. They may also make patient-specific dosimetry more accessible in institutions where computational resources or specialist expertise are limited.
However, AI-generated predictions remain sensitive to imaging artifacts, segmentation errors, scanner differences, incomplete training data, and changes in clinical practice. Models trained at one institution may not perform consistently in another without recalibration or external validation.
The broader significance is therefore not that AI will replace radiotheranostics, medical physics, or multidisciplinary clinical expertise. Its value lies in helping these disciplines operate with greater speed, consistency, and patient specificity. Used responsibly, AI can serve as a computational layer that strengthens precision treatment while preserving transparent validation, quality assurance, and human oversight.
Clinical Synergy: Integrating Y-90 with External-Beam and Molecular Imaging Modalities
One of the most promising directions in contemporary radiotheranostics is the integration of internal Yttrium-90 (Y-90) radioembolization with external-beam radiation techniques such as stereotactic body radiotherapy (SBRT) and external-beam radiation therapy (EBRT).
Y-90 microsphere distribution is governed by hepatic arterial anatomy, blood flow, catheter position, and tumor perfusion. As a result, the absorbed-dose distribution can be highly heterogeneous. Some regions may receive a substantial tumoricidal dose, while poorly perfused areas receive less radiation. These relatively underdosed regions are sometimes described as cold spots.
External-beam radiotherapy offers a complementary mechanism. Because it does not depend on intra-arterial microsphere deposition, EBRT or SBRT can be directed toward residual disease or regions that received insufficient internal radiation. In principle, this creates a combined strategy in which Y-90 delivers concentrated intra-arterial treatment and external radiation supplies a more spatially controlled dose to selected targets.
The clinical challenge is not simply to add the two treatments together. Their biological effects, spatial dose distributions, timing, treated liver volumes, and cumulative exposure to healthy tissue must be evaluated carefully. Effective integration therefore requires multimodality image registration, patient-specific dosimetry, and multidisciplinary treatment planning.
The Safety of Sequential Radiation
One of the principal concerns surrounding sequential Y-90 and external-beam treatment is cumulative liver toxicity. Patients undergoing radioembolization may already have cirrhosis, compromised functional liver reserve, previous systemic treatment, or extensive tumor burden. Additional radiation must therefore be planned around the amount and condition of the remaining healthy liver.
A 2026 retrospective study led by University of Cincinnati Cancer Center investigators reviewed 94 patients who received liver-directed EBRT, including 15 who had previously undergone Y-90 treatment. The investigators did not observe an increase in liver toxicity among the patients receiving EBRT after Y-90 and concluded that carefully individualized external radiation remained feasible in this selected population. The study provides meaningful reassurance, although its retrospective design and relatively small Y-90 subgroup mean that it should not be interpreted as establishing universal safety.
These findings support further investigation of individualized treatment sequences. Y-90 may be used first to achieve local control, reduce viable tumor volume, facilitate downstaging, or support transplantation strategies. Focused SBRT or EBRT may subsequently be considered for residual, recurrent, or insufficiently treated disease.
The decision must remain patient-specific. Important variables include baseline liver function, prior absorbed dose, treated liver volume, time between therapies, vascular anatomy, tumor location, and the dose delivered to uninvolved liver and adjacent organs.
Theranostic Pairs and Molecular Precision
Theranostics combines diagnostic imaging and targeted therapy through radiopharmaceuticals that share the same biological targeting mechanism. A diagnostic radionuclide is used to visualize and quantify target expression or biodistribution, while a therapeutic radionuclide delivers radiation to the same molecular target.
Some theranostic systems use different isotopes of the same element. Others use different radionuclides attached to the same or closely related targeting ligand. The objective is to determine whether a tumor expresses the intended target, estimate where the therapeutic compound is likely to accumulate, and support individualized activity selection and dosimetry.
The Y-86/Y-90 pair illustrates the matched-isotope concept. Y-86 is a positron-emitting radionuclide that can be imaged with PET, while Y-90 delivers therapeutic beta radiation. Preclinical research has demonstrated the use of Y-86 imaging to evaluate the biodistribution of Y-90-labeled targeting compounds. However, this approach remains distinct from routine Y-90 microsphere planning and should not be presented as an established component of liver radioembolization practice.
Radiotheranostic research is also expanding beyond established targets such as somatostatin receptors and prostate-specific membrane antigen. Fibroblast activation protein, or FAP, is being investigated because it is expressed by cancer-associated fibroblasts in the microenvironment of many solid tumors.
A 2024 preclinical study evaluated a Ga-68/Lu-177-labeled FAP-targeting compound known as FAPT. Ga-68 supported PET imaging, while Lu-177 provided therapeutic beta emission. The results demonstrated increased tumor uptake and prolonged retention in experimental models, but the compound remains investigational and should not yet be described as a broadly available clinical theranostic platform.
These developments illustrate the broader logic of molecular radiotheranostics: imaging can help determine whether a therapeutic target is present, quantify its distribution, and provide evidence for selecting or excluding a particular treatment.
Synergy with Immunotherapy
The interaction between radiation and the immune system is another important area of investigation. Radiation can cause tumor-cell injury, alter inflammatory signaling, release tumor-associated antigens, and reshape the local tumor microenvironment. These effects create a biological rationale for combining radiotherapy with immune checkpoint inhibition.
Both Y-90 radioembolization and SBRT can produce localized radiation effects within liver tumors. Researchers are investigating whether these effects can increase immune recognition or improve the activity of systemic immunotherapy in selected patients.
The EMERALD-Y90 study is evaluating transarterial radioembolization in combination with durvalumab and bevacizumab for hepatocellular carcinoma. The trial is designed to examine the safety and clinical activity of combining liver-directed radiation with immune checkpoint inhibition and antiangiogenic therapy. It should be described as an active clinical investigation rather than an established treatment pathway.
At present, it remains premature to conclude that Y-90 combined with immunotherapy improves overall survival. Clinical benefit must be demonstrated through appropriately designed prospective trials that distinguish the contribution of radioembolization from the effects of systemic therapy.
The most important questions include treatment sequencing, patient selection, radiation dose, liver function, immune-related toxicity, and whether local radiation can produce clinically meaningful systemic immune effects.
The Regulatory and Standardization Frontier
As artificial intelligence, advanced dosimetry, and new radiopharmaceutical combinations move toward clinical use, validation and standardization become increasingly important. A model that performs well in one retrospective dataset may not retain the same accuracy across different scanners, institutions, acquisition protocols, patient populations, or clinical workflows.
Innovation alone is therefore insufficient. AI-supported radiotherapy systems must also be reproducible, explainable within their intended context, continuously monitored, and subjected to appropriate quality assurance.
ESTRO–AAPM Guidelines for AIIn 2024, the European Society for Radiotherapy and Oncology and the American Association of Physicists in Medicine published a joint guideline addressing the development, clinical validation, and reporting of AI models in radiation therapy.
Developed through a Delphi consensus process, the guideline contains 19 statements covering major requirements for responsible AI development and clinical translation.
Its central priorities include:
Transparent reporting
The intended clinical use, model architecture, data sources, training procedures, validation methods, performance metrics, and known limitations should be described clearly.
Representative data
Development and validation datasets should reflect the clinical populations, imaging systems, and treatment conditions in which the model is intended to operate.
Independent validation
Performance should be evaluated outside the original development environment whenever possible. Internal validation alone cannot establish broad generalizability.
Clinical relevance
Technical accuracy must be connected to clinically meaningful outcomes. A statistically strong prediction is not automatically useful if it does not improve treatment decisions, workflow, safety, or patient outcomes.
Quality assurance and monitoring
AI systems require testing before deployment and continued surveillance after implementation. Performance can change as imaging equipment, clinical protocols, patient populations, or software environments evolve.
Together, these principles create a bridge between experimental model development and dependable clinical adoption.
Regulatory Engagement and FDA Guidance
In January 2025, the U.S. Food and Drug Administration issued draft guidance on the use of AI-generated information to support regulatory decision-making for drugs and biological products.
The guidance proposes a risk-based credibility framework tied to a model’s specific context of use. It asks developers to define the question being addressed, establish the intended role of the model, assess the risks associated with incorrect outputs, document credibility evidence, and determine whether model performance is sufficient for that particular application.
The scope of this guidance must be interpreted carefully. It primarily addresses AI used to produce information supporting regulatory decisions about drugs and biological products. It is not, by itself, a complete regulatory framework for every AI-enabled dosimetry or treatment-planning system.
Software used to calculate, recommend, or control patient treatment may be regulated as a medical device or as Software as a Medical Device, depending on its intended use and functionality. Such systems may be subject to separate device-specific requirements governing verification, validation, cybersecurity, software maintenance, human factors, and post-market monitoring.
The broader regulatory principle is nevertheless clear: the level of evidence should correspond to the clinical risk. A model used for research prioritization does not require the same safeguards as an algorithm whose output directly influences the radiation activity administered to a patient.
Infrastructure, Global Access, and Low- and Middle-Income Countries
The development of precision radiotherapy cannot be separated from the problem of unequal access. Advanced dosimetry and AI-supported planning have limited practical value in regions that lack basic imaging systems, radiotherapy equipment, trained professionals, reliable isotope supply, or sustainable maintenance programs.
The access gap remains substantial. The IAEA has reported that while radiotherapy is broadly available in many high-income countries, access is significantly lower in middle-income countries and may reach only a small proportion of patients in low-income settings.
Closing this gap requires more than donating equipment. Sustainable radiotherapy systems depend on trained clinicians, medical physicists, radiation therapists, radiochemists, engineers, regulatory capacity, quality-assurance programs, maintenance infrastructure, and long-term financial planning.
The IAEA Rays of Hope Initiative
The International Atomic Energy Agency launched the Rays of Hope initiative in February 2022 to support countries establishing or expanding medical imaging, nuclear medicine, and radiotherapy services. The program prioritizes high-impact interventions designed around each participating country’s needs and existing healthcare capacity.
Its work includes:
- supporting the acquisition and safe implementation of diagnostic and treatment equipment;
- strengthening professional education and clinical training;
- developing sustainable quality-assurance systems;
- supporting national cancer-control planning;
- and establishing Anchor Centres that provide regional expertise, education, and professional development.
A major milestone occurred in July 2025, when Malawi opened its first public radiotherapy center at Kamuzu Central Hospital in Lilongwe. The center was supported through the Rays of Hope initiative and expanded domestic access to treatment that had previously been extremely limited.
The Malawi center demonstrates that infrastructure development can produce measurable improvements in access. It also illustrates why equipment must be accompanied by workforce training, data collection, safety systems, and sustained technical support.
Decentralized Isotope Production and the Role of Accelerators
Medical isotope supply is another major component of global access. Many radionuclides depend on specialized reactors, generators, cyclotrons, linear accelerators, target-processing facilities, and international distribution networks. Short physical half-lives and complex radiochemistry make some products especially difficult to transport over long distances.
Accelerator-based production may help diversify supply for selected radionuclides. Medical cyclotrons are already essential for producing short-lived PET isotopes such as fluorine-18 and can support regional diagnostic imaging networks. Compact accelerator systems are also being investigated for the production of additional diagnostic and therapeutic radionuclides.
However, isotope production is highly radionuclide-specific. Cyclotrons and compact linear accelerators cannot be treated as universal replacements for nuclear reactors.
Production feasibility depends on:
- the required nuclear reaction;
- beam particle, energy, and current;
- target availability and cost;
- achievable radionuclide yield;
- radionuclidic purity;
- chemical separation requirements;
- shielding and waste management;
- target recovery and recycling;
- and the isotope’s physical half-life.
Y-90 deserves particular clarification. Clinical Y-90 is commonly obtained through the decay of its parent radionuclide, strontium-90, using generator-based production and specialized processing. A conventional hospital cyclotron or compact radiotherapy linac cannot simply replace this supply chain.
Accelerators can nevertheless strengthen the broader radiotheranostic ecosystem by supporting local or regional production of suitable PET, SPECT, alpha-emitting, or beta-emitting radionuclides where technically and economically feasible.
Production Infrastructure in Context
Nuclear reactors
Reactors remain important for producing many neutron-rich medical radionuclides and parent isotopes. They can provide high yields but require substantial national or regional infrastructure, regulatory oversight, waste management, and long-term maintenance.
Medical cyclotrons
Cyclotrons are well suited to producing many proton-rich diagnostic radionuclides, particularly short-lived PET isotopes. They can be installed at large hospitals or regional production centers but still require target systems, radiochemistry laboratories, quality-control capabilities, shielding, and trained personnel.
Linear accelerators
Specialized proton, ion, or electron linacs can produce selected medical radionuclides through particle- or photon-induced reactions. Their suitability depends on beam characteristics and the isotope being produced. Although they may offer advantages for certain production pathways, claims of universally lower cost, lower waste, or easier maintenance should be assessed on an isotope-specific basis.
A resilient global isotope strategy will likely use a distributed combination of reactors, generators, cyclotrons, linacs, processing centers, and regional distribution networks rather than relying on a single production technology.
Canada’s Contribution
Canada announced its first in-kind contributions to the IAEA’s global Rays of Hope activities in 2026 through work coordinated with the Canadian Nuclear Isotope Council.
The planned contribution includes an isotope-production and radiation-safety training program primarily hosted at Bruce Power. It is intended to provide regulators and specialists from low- and middle-income countries with practical exposure to isotope production, processing, and radiation-safety systems. Additional proposed clinical training includes radiochemistry, radiolabeling, quality control, dosimetry, medical imaging, and patient delivery.
The significance of this initiative extends beyond isotope supply. Long-term self-sufficiency requires regulatory knowledge, clinical expertise, production competence, safety culture, and the ability to maintain quality across the entire pathway from radionuclide generation to patient treatment.
Precision Radiopharmaceutical Medicine: The 2030 Vision
Looking toward the end of the decade, the convergence of radionuclide therapy, molecular imaging, external-beam radiation, AI-assisted analysis, and longitudinal clinical data points toward a more adaptive model of cancer treatment.
In this model, radiotherapy would no longer be planned as a series of isolated interventions. Instead, imaging, absorbed-dose measurements, laboratory biomarkers, treatment response, toxicity, and disease progression would be incorporated into a continuously updated decision process.
The goal would not be to automate oncology. It would be to give multidisciplinary teams a more complete and dynamic representation of the patient’s disease and the consequences of each treatment decision.
The Rise of the Digital Twin
A digital twin is more than a static three-dimensional reconstruction or a collection of patient data. In its strongest form, it is a patient-specific computational model that is repeatedly updated as new observations become available.
An oncology digital twin could potentially combine:
- anatomical and functional imaging;
- tumor segmentation and spatial dosimetry;
- pathology and molecular characteristics;
- circulating biomarkers such as circulating tumor DNA;
- laboratory indicators of liver function;
- previous treatment exposure;
- longitudinal response patterns;
- and uncertainty estimates associated with model predictions.
The system could then simulate possible treatment scenarios and compare their expected benefits and risks.
For example, a strong response to Y-90 radioembolization might support observation, systemic therapy, transplantation assessment, or a reduced external-beam treatment volume. Persistent viable disease or an underdosed region might support a focused SBRT boost, repeat locoregional treatment, or an alternative therapy.
This concept remains aspirational. Most current systems described as digital twins are more accurately characterized as multimodal predictive models or longitudinal digital representations. A clinically meaningful digital twin would require repeated calibration, prospective validation, uncertainty quantification, and evidence that twin-informed decisions improve outcomes.
Until those requirements are met, the more precise term is digital-twin-inspired treatment planning.
Integrating Biology at a Finer Scale
Future treatment planning may also incorporate biological features that are not visible through anatomy alone. Tumors frequently contain regions with different levels of perfusion, hypoxia, cellular density, metabolism, immune activity, and treatment resistance.
Multiparametric imaging and AI-supported analysis may help identify these subregions. Instead of prescribing a uniform dose to the entire visible tumor, clinicians could potentially increase radiation to resistant areas while reducing exposure to sensitive normal tissue.
This strategy is often described as biological dose painting.
For biological dose painting to become clinically dependable, imaging biomarkers must be reproducible and linked to meaningful biological characteristics. Treatment-planning systems must also account for spatial uncertainty, motion, changes between imaging and treatment, and the possibility that tumor biology evolves over time.
The objective is not simply greater technical resolution. It is to connect measurable biological differences with treatment decisions that produce better outcomes.
Toward Long-Term Disease Control
The broader objective of adaptive radiotheranostics is not to maximize radiation at every opportunity. It is to achieve durable disease control while preserving liver function, limiting toxicity, and maintaining quality of life.
This may involve different treatment combinations for different patients:
- radioembolization for concentrated intra-arterial treatment;
- external-beam radiation for spatially defined residual disease;
- molecularly targeted radiopharmaceuticals for tumors expressing suitable targets;
- systemic therapy for extrahepatic or biologically aggressive disease;
- and transplantation, surgery, or ablation when clinically appropriate.
The future of precision radiation medicine will therefore depend less on a single dominant technology and more on the ability to coordinate multiple therapies around the evolving condition of the patient.
Synthesis and Strategic Recommendations
The next generation of radiopharmaceutical medicine will depend on the convergence of radiation physics, molecular targeting, computational modeling, clinical expertise, and sustainable healthcare infrastructure.
Several priorities stand out.
Expand Multi-Institutional Data Sharing
AI models require large, diverse, and well-characterized datasets. Multi-institutional collaboration is essential for determining whether a model performs reliably across different populations, scanners, microsphere platforms, imaging protocols, and treatment practices.
Data-sharing initiatives should include standardized definitions, imaging metadata, segmentation protocols, absorbed-dose calculations, treatment parameters, clinical outcomes, and toxicity reporting.
Privacy-preserving approaches, federated learning, and common data models may help institutions collaborate without transferring all patient-level data into a single centralized repository.
Validate Combined-Modality Treatment Prospectively
Retrospective evidence suggests that selected patients can receive external-beam radiation after Y-90 without unacceptable toxicity, but larger prospective studies are needed.
Future trials should define:
- cumulative liver-dose constraints;
- optimal treatment intervals;
- appropriate patient-selection criteria;
- the role of SBRT for residual or underdosed disease;
- and the interaction between radioembolization, external radiation, and systemic therapy.
Randomized or carefully controlled studies will be especially important when evaluating combinations involving immunotherapy.
Require Transparent AI Validation
AI dosimetry systems should be evaluated against appropriate physics-based reference methods and tested outside their original development institutions.
Validation should include technical accuracy, clinical relevance, failure detection, subgroup performance, workflow integration, and post-deployment monitoring.
Models should communicate uncertainty rather than presenting every prediction as equally reliable.
Build Isotope-Specific Supply Strategies
Infrastructure planning should be based on the radionuclides a region intends to produce and use. Reactor, generator, cyclotron, and accelerator systems should not be compared only through broad claims about cost or safety.
Each production route requires analysis of yield, purity, target supply, radiochemistry, waste, workforce, distribution distance, and clinical demand.
Regional production hubs may provide a more realistic model than attempting to place every production technology within every hospital.
Strengthen Global Training Standards
As radiotherapy becomes more computationally intensive, education must extend beyond isotope administration and equipment operation.
Clinicians, physicists, radiochemists, technologists, regulators, and engineers will need training in:
- patient-specific dosimetry;
- AI model interpretation;
- data quality;
- uncertainty analysis;
- software validation;
- radiation safety;
- and continuous quality assurance.
Frameworks such as the ESTRO–AAPM guideline and the IAEA Rays of Hope Anchor Centres provide foundations for building this expertise internationally.
ConclusionBy the early 2030s, the most advanced radiotherapy pathways may combine molecular targeting, patient-specific dosimetry, external-beam precision, longitudinal imaging, and computational decision support within one adaptive model of care.
The deeper goal is not technological complexity for its own sake. It is to deliver the appropriate radiation to the appropriate target at the appropriate time while preserving healthy tissue, maintaining clinical accountability, and extending access across very different healthcare systems.
The future of radiotheranostics will be defined not only by what can be calculated or delivered, but by whether those capabilities can be validated, governed, and made meaningfully available to patients.
Recommended Reading
Combined Radiation and Immunotherapy
- University of Cincinnati Cancer Center report on EBRT following Y-90 treatment.
- EMERALD-Y90 clinical trial record: transarterial radioembolization with durvalumab and bevacizumab.
Theranostic Pairs
- 86/90Y-Based Theranostics Targeting Angiogenesis in a Murine Breast Cancer Model.
- 68Ga/177Lu-Labeled Theranostic Pair for Targeting Fibroblast Activation Protein with Improved Tumor Uptake and Retention.
AI Validation and Regulation
- A Joint ESTRO and AAPM Guideline for Development, Clinical Validation and Reporting of Artificial Intelligence Models in Radiation Therapy.
- FDA draft guidance, Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products.
Global Access and Infrastructure
- IAEA Rays of Hope initiative.
- IAEA reporting on Malawi’s first public radiotherapy center.
- Production Review of Accelerator-Based Medical Isotopes.
- Canadian Nuclear Isotope Council and IAEA training initiative for low- and middle-income countries.