AI has rapidly evolved from a theoretical concept into a practical clinical tool, with radiation oncology emerging as one of the medical specialties most prepared to benefit from its integration. Radiotherapy is inherently dependent on advanced imaging, computational modeling, and precise dose delivery, all of which generate large and complex datasets.
This technological foundation has made radiotherapy particularly receptive to artificial intelligence, especially machine learning and deep learning approaches that can analyze data at a scale and speed beyond human capacity. As cancer incidence rises globally and workforce pressures increase, AI is increasingly viewed not as an optional enhancement but as a strategic enabler of efficient, standardized, and high-quality radiotherapy delivery.
The role of AI in radiation oncology is not to replace clinical expertise, but to complement it. By automating repetitive and time-consuming processes, AI allows clinicians, physicists, and dosimetrists to focus on tasks that require judgment, experience, and patient-centered decision-making. This collaborative model of human–AI interaction is shaping the modern radiotherapy workflow.
Integration of AI Across the Radiotherapy Workflow
Radiotherapy planning and delivery consist of multiple interconnected stages, each of which influences treatment accuracy and safety. AI applications are increasingly present from the earliest phases of image acquisition to treatment adaptation and quality assurance. In imaging, deep learning-based reconstruction and denoising techniques have been investigated to improve image quality and reduce artifacts, particularly in lower-dose acquisitions. Higher-quality images support more reliable contouring and dose calculations downstream, strengthening the overall planning process.
One of the most mature and widely adopted applications of AI is auto-segmentation. Manual delineation of target volumes and organs at risk is labor-intensive and subject to interobserver variability, which can affect both efficiency and plan consistency. Deep learning-based contouring tools have demonstrated the ability to significantly reduce contouring time while improving standardization across disease sites. These systems perform particularly well for organs at risk, although clinician review and editing remain essential to ensure clinical accuracy and appropriateness.
AI-assisted treatment planning has also gained momentum. Machine learning models are being explored for dose prediction, plan optimization, and rapid generation of clinically acceptable treatment plans. By reducing the need for repeated manual adjustments, these tools can shorten planning timelines and support more reproducible plan quality across planners and institutions. This is especially relevant for complex treatment techniques, where planning traditionally requires significant time and expertise.
Radiation Therapy Toxicity Myth-Busting
Adaptive radiotherapy represents another area where AI shows promise. Adaptive workflows depend on frequent imaging and rapid reassessment of anatomy to determine whether treatment plans should be modified.
AI can accelerate this process through faster segmentation, improved image registration, and automated identification of clinically meaningful anatomical changes. Given the time-sensitive nature of adaptive decision-making, the integration of AI in this context must be accompanied by robust safety mechanisms and clearly defined clinical oversight.
Beyond planning and adaptation, AI is increasingly discussed in the context of quality assurance and workflow reliability. Machine learning models trained on historical planning and delivery data can help identify anomalies, flag outliers, and support early error detection. These tools are designed to complement established physics-based QA processes, adding an additional layer of surveillance in increasingly complex radiotherapy environments.
Workflow Optimization and Clinical Efficiency
In practical terms, workflow optimization refers to reducing delays, minimizing unnecessary manual effort, and creating more predictable planning and treatment timelines without compromising quality.
AI contributes to these goals by decreasing the time required for contouring, streamlining planning iterations, and supporting consistent application of clinical constraints. These improvements are particularly valuable in high-volume centers and in health systems facing staffing shortages, where inefficiencies can translate directly into treatment delays and increased patient anxiety.
However, efficiency gains alone are insufficient. The clinical value of AI-enabled workflows must be assessed in terms of edit burden, dosimetric impact, and overall plan acceptability. Experience from early adopters suggests that the most successful implementations are those that integrate AI outputs seamlessly into existing workflows while preserving clear checkpoints for human review and decision-making.
Multimodal and Context-Aware AI Systems
An important emerging direction is the development of multimodal AI systems that integrate imaging data with clinical information. Target delineation and treatment planning often depend on factors that extend beyond anatomy visible on a planning scan, such as tumor stage, surgical history, and treatment intent.
Recent studies have explored large language model-driven systems that combine clinical text with imaging features to enhance contouring accuracy and generalizability. These approaches reflect a broader shift toward context-aware AI tools that more closely align with real-world clinical reasoning.
Re-irradiation Strategies: Patient Selection and Safety Considerations, Reirradiation (reRT)
Validation, Standards, and Governance
The safe and effective use of AI in radiotherapy requires rigorous validation and structured governance. AI performance can vary substantially across institutions due to differences in imaging protocols, contouring practices, and patient populations.
As a result, professional guidance increasingly emphasizes clear definition of intended use, transparent reporting, local commissioning, and continuous performance monitoring. A joint ESTRO and AAPM guideline addresses the development, clinical validation, and reporting of AI models in radiation therapy, underscoring the need for standardized evaluation frameworks.
Expert and cooperative group reviews further highlight the importance of integrating AI tools thoughtfully into clinical workflows, with attention to failure modes that may only emerge during real-world use. These considerations reinforce the principle that AI should be treated as a clinical technology requiring the same level of scrutiny as other radiotherapy systems.
Ongoing Challenges
Despite rapid progress, several challenges continue to shape the adoption of AI in radiotherapy. High-quality, representative datasets are essential for robust model development, yet clinical data remain heterogeneous and difficult to aggregate across institutions. Privacy and data governance requirements add further complexity, particularly for multi-institutional collaboration.
Trust and explainability also influence clinical acceptance. When AI systems operate as “black boxes,” clinicians may struggle to understand or anticipate errors, highlighting the importance of transparency, structured QA, and continuous oversight. Additionally, while AI clearly improves efficiency and standardization, demonstrating direct improvements in patient outcomes remains an important focus for future research.
Written by Nare Hovhannisyan, MD