Adaptive radiation therapy — the practice of modifying the treatment plan during a course of fractionated radiotherapy to account for anatomical changes in the patient — has been a subject of active clinical research for nearly two decades. The case for adaptation is biologically clear: bladder filling varies by up to 100 mL fraction-to-fraction in prostate patients; cervical cancers shrink measurably over a six-week course; head-and-neck patients lose parotid volume and weight as treatment progresses. Plans optimized on the simulation CT are, to varying degrees, physically delivering different dose distributions by week four or five.
In Japan, the clinical uptake of adaptive RT follows a distinct pattern shaped by equipment infrastructure, physicist staffing models, and clinical workflow conventions that differ meaningfully from the US and European contexts where most published adaptive RT protocols originated.
The Japanese Clinical Infrastructure Context
Japan has one of the highest per-capita linear accelerator installation densities among high-income countries, with equipment installed across university hospitals, cancer centers, and regional community hospitals. However, the physicist-to-linac ratio in many of these facilities is tighter than in large North American academic departments. A department running four linacs with two full-time clinical physicists has a different capacity for adaptive workflow implementation than a department with the same equipment and six physicists.
This staffing constraint shapes which adaptive RT approaches are feasible in practice. Online adaptive RT — which requires same-fraction re-planning, typically within fifteen to thirty minutes while the patient remains on the treatment couch — demands physicist availability and computing resources that not all departments can sustain for high-volume protocols. Offline adaptive RT, where re-planning decisions are made between fractions based on accumulated imaging, is more compatible with existing staffing models but provides less individual-fraction accuracy.
The distinction between online and offline adaptive RT is not merely technical — it determines which patient populations benefit, which equipment configurations are required, and how physicist workflow is restructured. Most Japanese departments currently implementing adaptive RT are doing so in offline or hybrid modes, with online approaches concentrated at high-volume academic centers.
Imaging Infrastructure and Plan-of-the-Day Workflows
The plan-of-the-day (POTD) approach — generating a library of pre-optimized plans during simulation, then selecting the best-matching plan at each fraction based on on-couch imaging — represents an intermediate adaptive strategy that has found practical traction in Japan for bladder and prostate protocols. The approach requires CBCT-based anatomy matching at each fraction and a plan library that adequately covers the range of observed anatomical states, but it avoids the re-optimization step that makes true online adaptive RT staffing-intensive.
For bladder IMRT, where filling state is the dominant source of interfractional anatomy change, POTD libraries of three to five plans covering empty-to-full bladder configurations can capture a substantial fraction of the clinically relevant anatomy range. At a university hospital radiation department in the Chubu region running approximately twenty bladder IMRT courses annually, POTD implementation reduced estimated out-of-field bladder dose compared to a single fixed plan — without requiring physicist presence for each fraction selection, because the protocol defines the selection rule and the radiation therapist applies it under periodic physicist oversight.
Where Anatomy Deformation Creates Technical Barriers
For true re-optimization-based adaptive RT — whether online or inter-fraction — deformable image registration (DIR) is the enabling technology that maps anatomy from one image to another and propagates dose-volume information across the course of treatment. The clinical accuracy of DIR in regions with large deformation (sigmoid colon, bowel loops, post-surgical soft tissue) remains a source of clinical physics concern in Japan, as elsewhere.
Japan's clinical physics community has been cautious about adopting DIR-dependent adaptive workflows without institution-specific validation. The JASTRO (Japan Society for Therapeutic Radiology and Oncology) technical guidelines on adaptive RT emphasize validation of DIR accuracy before clinical implementation, and several academic departments have published DIR validation work for specific anatomical sites. This caution is scientifically appropriate — DIR errors propagate into adaptive plan quality assessments in ways that are not always visible in standard QA.
We are not suggesting that DIR technology is insufficiently mature for clinical use — it clearly is in use clinically at many institutions with appropriate protocols. The point is that the validation burden before clinical implementation is real, and departments that have not performed site-specific DIR validation should not assume accuracy from vendor-reported metrics alone.
MR-Linac and the Adaptive RT Frontier
MR-guided radiation therapy, which enables on-couch soft-tissue imaging with superior contrast to CBCT and supports same-fraction re-optimization, represents the current frontier of online adaptive RT. The number of MR-linac installations in Japan remains small relative to total linac installations, with systems installed at a limited number of high-volume academic centers as of mid-2025. The clinical workflows for MR-linac adaptive RT — online contouring, online re-optimization, online dosimetric QA — are substantially more demanding than CBCT-based approaches and require dedicated physicist resource allocation at each fraction.
For the majority of Japanese radiation oncology departments, MR-linac adaptive RT is a reference benchmark rather than an immediate implementation target. The more practically relevant near-term development is the maturation of AI-assisted re-planning tools that reduce the time cost of offline adaptive approaches on standard CBCT-equipped linacs — bringing meaningful adaptation within reach of departments that cannot support online re-optimization staffing models.
The Role of AI in Making Adaptive RT Accessible
The key bottleneck in adaptive RT, independent of the imaging modality, is re-planning speed and physicist workload. For an offline adaptive protocol where re-planning is triggered at fractions five and ten based on accumulated anatomy change, the department needs to complete a full re-plan — contour propagation from deformable registration, re-optimization against the updated anatomy, QA of the adapted plan — within a clinical day, without displacing the physicist's other work.
AI-assisted re-planning compresses the optimization iteration step, which is typically the largest time component after deformable registration. An AI planning system trained on the department's historical plan patterns can generate an initial adapted plan from the re-registered anatomy in minutes rather than the thirty to sixty minutes required for manual re-optimization from a blank starting state. The physicist's role shifts to evaluating and approving the AI-generated adapted plan rather than generating it from scratch.
This workflow model — AI-generated initial adapted plan, physicist review and approval — is consistent with how adaptive RT implementation is being approached at growing clinical centers across Japan. The technology reduces the per-fraction physicist time cost enough to make offline adaptive RT viable in departments where it would otherwise exceed staffing capacity. It does not remove the physicist from the adaptive RT workflow; it makes the physicist's participation in that workflow sustainable.