Xun Jia

Artificial intelligence (AI) has the potential to address workforce shortages and workflow inefficiencies in radiation oncology, particularly in low-resourced settings where limited specialist capacity constrains access to care. This systematic review evaluated AI deployments in radiation oncology in low- and middle-income countries (LMICs) and assessed readiness for safe implementation and scale-up. Following PRISMA 2020 guidelines, six databases were searched from January 2000 through December 2025. Eligible studies were classified using a radiation oncology-specific integration spectrum (Levels 0-4) and evaluated for workflow integration, validation, deployment, and governance characteristics. Eighteen studies met the inclusion criteria. Two studies (11.1%) focused on readiness or governance without clinical AI deployment (Level 0), nine (50.0%) reported task-level applications (Level 1), six (33.3%) demonstrated clinician-supervised workflow integration (Level 2), and one (5.6%) reported cross-stage workflow orchestration (Level 3). No study provided sufficient evidence for classification as Level 4 operation. Validation was predominantly retrospective, and continuous performance monitoring was not reported. AI deployment in LMIC radiation oncology remains concentrated at early stages of integration. This pattern reflects not a lack of innovation, but constraints in validation, workflow maturity, infrastructure, and governance. Safe and scalable deployment will require coordinated advances across these domains.

Background: Whole‐brain radiotherapy (WBRT) is a common treatment due to its simplicity and effectiveness. While automated Field‐in‐Field (Auto‐FiF) functions assist WBRT planning in modern treatment planning systems, it still requires manual approaches for optimal plan generation including patient‐specific hyperparameters definition and plan refinement based on quality feedback. Purpose: This study introduces an automated WBRT planning pipeline that integrates a deep learning (DL) Hyperparameter Prediction model for patient‐specific parameter generation and a large‐language model (LLM)‐based conversational interface for interactive plan refinement. Methods: The Hyperparameter Prediction module was trained on 55 WBRT cases using geometric features of clinical target volume (CTV) and organs at risk (OARs) to determine optimal Auto‐FiF settings in RayStation treatment planning system. Plans were generated under predicted hyperparameters. For cases in which the generated plan was suboptimal, quality feedback via voice input was captured by a Conversation module, transcribed using Whisper, and interpreted by GPT‐4o to adjust planning settings. Plan quality was evaluated in 15 independent cases using clinical metrics and expert review, and model explainability was supported through analysis of feature importance. Results: Fourteen of 15 DL‐generated plans were clinically acceptable. Normalized to identical CTV D95% as the clinical plans, the DL‐generated and clinical plans showed no statistically significant differences in doses to the eyes, lenses, or CTV dose metrics D1% and D99%. The DL‐based planning required under 1 min of computation and achieved total workflow execution in approximately 7 min with a single mouse click, compared to 15 min for manual planning. In cases requiring adjustment, the Conversational module successfully improved dose conformity and hotspot reduction. Conclusions: The proposed system improves planning efficiency while maintaining clinically acceptable plan quality. It demonstrates the feasibility of combining DL‐based hyperparameter prediction with LLM interaction for streamlined, high‐quality WBRT planning.

Objective: Biological effect of charged particles characterized by cell surviving fraction (SF) depends on physical factors such as particle type, dose, beam quality descriptors, e.g. linear energy transfer (LET) or lineal energy. Driven by the crucial role of DNA double-strand breaks (DSBs) as indicators of cellular responses to ionizing radiation, this study aims to develop a cell-specific SF model based on the initial DSBs of varying complexities universal with respect to photon, proton, and carbon-ion irradiations across different beam qualities. Approach: Microscopic Monte Carlo simulations (MMCSs) were used to calculate DSBs for experiments measuring SF of H460 and H1437 lung cancer cells irradiated by proton (dose-mean lineal energy y ̅_d 2.0-20.0 keV/ µm), carbon ions (y ̅_d 18.6-87.9 keV/µm) and reference 137Cs photon beam. We modeled SF as a third-order polynomial function of DSBs of various complexities with two, three, and more strand breaks. The function was determined by a data fitting process. Main results: For each cell line, a single model was able to accurately describe SF in experiments under all 19 irradiation conditions with different particle types and y ̅_d values. Root Mean Square Errors were 0.218 for H460 cells and 0.170 for H1437 cells. When applying the model to a proton spread-out-Bragg-peak case, calculated relative biological effectiveness at SF=0.1 were 1.01-1.42 for the H460 cells and 1.03-1.65 for the H1437 cells depending on the depth. Significance: We successfully developed a SF model using DSBs of various complexities as input variables. The model is universally applicable to photon, proton, and carbon-ion irradiations across different beam qualities.

Bone densitometry in osteoporosis diagnosis via dual-energy X-ray absorptiometry (DXA) can benefit from advances in imaging detector technology. We devised a compact imaging scanner—DXA4A—using a photon-counting and energy-sensitive Timepix4 hybrid pixel detector (512 × 448 pixels, 55 µm pitch), for areal bone mineral density (aBMD) assessments in the distal radius and tibia in the clinic and for future in-flight astronauts’ bone health assessment. We present the design and Monte Carlo simulations of the scanner. A Timepix4 detector with a 1 mm thick CdTe sensor was tested in the laboratory with X-ray tube sources, acquiring first images of test samples. Monte Carlo simulations were implemented for scanner design and performance prediction, using 50 kVp unfiltered and 100 kVp Sm K-edge filtered spectra. With a digital twin of the scanner and patient wrist, we set up a virtual imaging study and determined the aBMD in the forearm of a patient (0.515 ± 0.048 g/cm²), in agreement with the clinical DXA value (0.571 g/cm² for the total forearm). This study highlights the feasibility of realizing a compact DXA scanner for the distal tibia and radius with spectral capabilities, exploiting Timepix4 hybrid detectors for its peculiar energy sensitivity and photon event timing properties for tissue identification.

This roadmap provides a comprehensive framework for integrating diffusion-weighted imaging (DWI) into radiation therapy (RT), with an emphasis on its application in magnetic resonance imaging-guided radiotherapy (MRIgRT) and its potential for driving biological image-guided adaptive radiotherapy (BIgART). Developed through collaboration among experts in medical physics, MRI imaging science, and radiation oncology, the paper aims to bridge disciplinary gaps and foster a shared understanding across scientific, technical, and clinical domains. It benchmarks the current state of DWI in RT, identifies critical challenges, and highlights recent advancements in acquisition, reconstruction, biophysical modeling, quality assurance, clinical validation and translation, as well as emerging concepts. By outlining ongoing efforts and forecasting future developments, this roadmap supports the adoption of DWI as a quantitative imaging biomarker for personalized and adaptive radiotherapy in precision oncology.

Objective. Microcalcification ( µ Calc) detection plays an important role in breast cancer screening. Electronic noise in energy-integrating detectors (EIDs) is the major challenge for this task in current breast cone-beam CT (bCBCT) due to the tight dose constraint for breast imaging. bCBCT with a photon counting detector (PCD) can potentially offer a higher spatial resolution and lower noise. This study performed a direct comparison of bCBCTs with the two detector types via GPU-based Monte Carlo (MC) simulation. Approach. We employed Virtual Clinical Trial for Regulatory Evaluation toolkit to generate a realistic breast phantom with a 0.25 mm 3 voxel size, 80% fat fraction and 14 cm diameter. We considered a bCBCT system with a 60 kV x-ray source filtered with 0.3 mm Cu and detector response functions for PCD and EID. A total of 360 projections were simulated with a total number of 3.15 × 10 12 photons, corresponding to ∼4 mGy mean glandular dose, comparable to a two-view mammography. We modified our GPU-based MC simulation code to incorporate analytical descriptions of µ Calcs of spherical shapes with diameters ranging from 0.1 to 0.4 mm, in 0.1 mm increments, into the voxelized phantom. A nichrome wire with 0.07 mm diameter was simulated to calculate the modulation transfer functions (MTFs). bCBCT images were reconstructed with the Feldkamp–Davis–Kress algorithm, and image quality and µ Calc detection performance were evaluated. Main results. EID-bCBCT had more profound image noise due to electronic noise. The image intensity standard deviations estimated within a region of interest were 0.055 cm ⁻¹ for EID-bCBCT and 0.038 cm ⁻¹ for PCD-bCBCT, respectively. µ Calcs and breast anatomy such as ligaments were more visible in the PCD-bCBCT images. The 10% MTF cutoffs were 5.5 and 9.5 lp mm ⁻¹ for EID-bCBCT and PCD-bCBCT, respectively. Contrast-to-noise ratio ranged in 1.20–9.13 for EID-bCBCT and 3.07–14.74 for PCD-bCBCT, depending on µ Calc sizes. Significance. We compared EID- and PCD-based bCBCT for µ Calc detection using GPU-based MC simulations in a clinically realistic setting. Our results demonstrate a potential advantage of PCD-bCBCT for this detection task.

Current radiation therapy treatment planning is limited by suboptimal plan quality, inefficiency, and high costs. This perspective paper explores the complexity of treatment planning and introduces Human-Centric Intelligent Treatment Planning (HCITP), an AI-driven framework under human oversight, which integrates clinical guidelines, automates plan generation, and enables direct interaction with planners. We expect that HCITP will enhance efficiency, potentially reducing planning time to minutes, and will deliver personalized, high-quality plans. Challenges and potential solutions are discussed.

Objective. This study aims to uncover the opaque decision-making process of an artificial intelligence (AI) agent for automatic treatment planning. Approach. We examined a previously developed AI agent based on the actor-critic with experience replay (ACER) network, which automatically tunes treatment planning parameters (TPPs) for inverse planning in prostate cancer intensity modulated radiotherapy. We selected multiple checkpoint ACER agents from different stages of training and applied an explainable AI method to analyze the attribution from dose-volume histogram (DVH) inputs to TPP-tuning decisions. We then assessed each agent’s planning efficacy and efficiency, and evaluated their policy space and final TPP tuning space. Combining findings from these approaches, we systematically examined how ACER agents generated high-quality treatment plans in response to different DVH inputs. Main results. Attribution analysis revealed that ACER agents progressively learned to identify dose-violation regions from DVH inputs and promote appropriate TPP-tuning actions to mitigate them. Organ-wise similarities between DVH attributions and dose-violation reductions ranged from 0.25 to 0.5 across tested agents. While all agents achieved comparably high final planning scores, their planning efficiency and stability differed. Agents with stronger attribution-violation similarity required fewer tuning steps ( 12–13 vs 22), exhibited a more concentrated TPP-tuning space with lower entropy ( 0.3 vs 0.6), converged on adjusting only a few key TPPs, and showed smaller discrepancies between practical tuning steps and the theoretical steps needed to move from initial values to the final TPP space. Putting together, these findings indicate that high-performing ACER agents can effectively identify dose violations from DVH inputs and employ a global tuning strategy to achieve high-quality treatment planning. Significance. This study demonstrates that the AI agent learns effective TPP-tuning strategies, exhibiting behaviors similar to those of experienced human planners. Improved interpretability of the agent’s decision-making process may enhance clinician trust and inspire new strategies for automatic treatment planning.