Paul Keall

Objective. Real-time six-degree-of-freedom (6DoF) tumour motion monitoring is important for accurate stereotactic body radiotherapy (SBRT) of thoracic and abdominal cancer sites, yet existing approaches either require specialised imaging hardware or continuous fluoroscopic imaging. This study presents the development and first experimental implementation of a real-time image-guided 6DoF internal-external correlation (6D-IEC) modelling framework on a standard linear accelerator. Approach. The 6D-IEC framework estimates 6DoF tumour motion using a state-augmented linear correlation model between a 1DoF external respiratory monitor and internal fiducial marker positions, updated with intrafraction kV images every 3 s. Performance was characterised through two complementary evaluations: a prospective experimental investigation on a robotic motion platform, providing independent ground truth across eight patient-measured lung and liver motion traces (3 regular, 5 irregular); and a patient data study using intrafraction imaging data from three patients treated in the TROG 17.03 LARK liver SBRT trial (NCT02984566). 6DoF geometric error was quantified for both evaluations; software processing latency was measured in the experimental investigation. Main results. In the experimental investigation, mean ± SD geometric errors were 0.1 ± 1.0 mm, 0.3 ± 1.9 mm, and −0.3 ± 1.4 mm along the left–right (LR), superior–inferior (SI), and anterior–posterior (AP) translational axes, and 0.4°± 1.2°, −0.8°± 1.0°, and −0.2°± 0.8° around the corresponding rotational axes (rLR, rSI, rAP). Software processing latency was 190 ± 80 ms. In the patient data study, errors were −0.5 ± 0.7 mm, −1.2 ± 2.5 mm, −0.1 ± 1.0 mm (LR, SI, AP), and 1.1°± 1.7°, −0.9°± 1.5°, and −0.4°± 1.6° (rLR, rSI, rAP). Significance. This work demonstrates that real-time 6DoF motion monitoring with clinically relevant accuracy is achievable on a standard linear accelerator without specialised hardware or continuous kV imaging. 6D-IEC infers tumour motion from an external signal between imaging updates, with accuracy contingent on IEC, but requires approximately 95% fewer intrafraction kV images than continuous fluoroscopic approaches. This has implications for broader access to real-time image-guided radiotherapy.

Background: Oral mucositis is a debilitating yet potentially modifiable toxicity frequently encountered in patients receiving head and neck radiotherapy. This study aimed to evaluate mid-treatment functional changes in the oral cavity using fluorodeoxyglucose positron emission tomography/computed tomography (FDG-PET/CT) and correlate these early imaging findings with moderate-severe mucositis at the completion of radiotherapy in patients with mucosal head and neck squamous cell carcinoma (HNSCC). Methods: A total of 54 patients with newly diagnosed mucosal HNSCC without distant metastases from two prospective imaging biomarker studies received FDG-PET/CT imaging at baseline and mid-treatment (week 3). The oral cavity was volumetrically segmented at baseline on the CT component of the CT-PET, with this volume transferred to mid-treatment imaging. The median standardised uptake value (SUV_median) on FDG-PET was measured at these timepoints. Absolute values and relative change (Δ) in SUV_median were correlated with the development of moderate-to-severe mucositis at treatment completion. Four predictive models, including a reference clinical model, were developed using multivariable logistic regression incorporating clinical and radiotherapy dosimetric parameters. Results: Forty-six patients (82.1%) developed moderate-severe [Common Terminology Criteria for Adverse Events (CTCAE) grade ≥2] mucositis at completion of treatment. The average oral cavity SUV_median (g/mL) increased from baseline to mid-treatment. Change in oral cavity SUV_median from baseline to mid-treatment was associated with mucositis (CTCAE grade ≥2) on univariate (P=0.04) and multivariate analysis (P=0.049). The reference ‘clinical model’ correlated with mucositis [area under the curve (AUC) 0.784, P=0.003]. Addition of oral cavity ΔSUV_median to the clinical model resulted in the highest correlation to mucositis at completion of treatment (AUC 0.855, PConclusions: Our study demonstrated that functional changes in the oral cavity as seen on mid-treatment FDG-PET/CT correlate with development of moderate-severe mucositis at completion of radiotherapy. Integration of this imaging data with clinical variables may enhance mucositis risk prediction, with potential application in the individualisation of head and neck radiotherapy.

N/A - No abstract for viewpoint articles

Background: Following the release in 2016 of the report of the American Association of Physicists in Medicine Task Group 100, there has been growing interest in the use of prospective hazard analysis in radiation therapy. System Theoretic Process Analysis (STPA) is an emerging technique in this domain that is particularly suited to processes that involve time sensitive collaboration, decision‐making and/or automation. Purpose: The goal of this research was to use STPA to evaluate existing processes and procedures with an aim to identify improvements, gaps or unforeseen risks stemming from implementing real‐time adaptive treatment on a helical tomotherapy platform. Methods: The Radixact treatment delivery system (Accuray Inc., Sunnyvale, CA, USA), an evolution of the Tomotherapy platform, incorporates upgrades such as the Synchrony system for real‐time motion monitoring and treatment adaptation. In collaboration with a team from the radiation oncology department of a large public hospital, a prospective hazard analysis focused on the real‐time adaptive capabilities of the Radixact Synchrony system was conducted using STPA. The system boundaries were defined and a control structure model comprising sub‐systems and control actions was developed. Unsafe control actions were identified and broad‐based causal scenarios were generated. The causal scenarios that were novel, specific to Synchrony or challenging to mitigate were selected for further analysis regarding impacts and potential causes, following which mitigation strategies were proposed, taking into consideration the hierarchy of controls. Results: A control structure model encompassing the entire patient journey was developed, incorporating all the hardware and software components and human decision makers. The model consisted of 12 sub‐systems and 21 control actions, resulting in 108 unsafe control actions and 595 causal scenarios. Sixty‐one causal scenarios were selected for further analysis, for which mitigation strategies were proposed based on the hierarchy of controls. These included the development of better reference documentation, the systematic testing of the sensitivity of tracking performance to changes in tracking parameters, guidance around setting and documenting tracking parameters, and documentation review. Conclusions: STPA was effectively used to assess the Radixact Synchrony system's real‐time adaptive radiation therapy capabilities, providing insight into how the system could become unsafe throughout the patient journey. While focused on Radixact Synchrony and real‐time adaptive radiation therapy, this study offers a transferable example of STPA application, from analysis initialization to mitigation, that can inform other safety assessments in radiation therapy.

Objective: Accurate dose delivery in the presence of anatomical motion and deformation remains a major challenge in radiation therapy. Real-time dose-guided radiation therapy addresses this challenge by integrating volumetric imaging, dose accumulation, and adaptive treatment in a quasi-continuous manner. Unlike previous studies limited to rigid motion or phantom data, this study employs a motion model to generate real-time volumetric images, enabling deformable dose-guided multileaf collimator (MLC) tracking in lung and liver cancer patients. Approach: Deformable dose-guided MLC tracking was simulated for ten patients (4 lung, 6 liver). The workflow included: (1) training a patient-specific linear regression-based motion model relating external respiratory signals (RPM) to deformation vector fields using planning 4DCT data, (2) generating volumetric images using treatment RPM signals, (3) simulating dose-guided MLC tracking comparing deformable, rigid and no tracking, and (4) evaluating the motion model accuracy using fiducial markers and kV imaging. Dosimetric accuracy was assessed using 3D Gamma analysis (2%/2 mm) and normalized root mean squared error by comparing simulated delivered dose to a motion-compensated baseline. Motion model error was quantified as the distance between predicted and ground truth marker positions. Main results: Deformable tracking significantly outperformed rigid and no tracking (p < 0.001), achieving a mean Gamma passing rate of 95.0 ± 6.0%, compared to 86.3 ± 10.0% and 74.8 ± 15.2%, respectively. Lung patients showed a greater benefit in Gamma passing rate (deformable: 98% vs. rigid: 85%, +13%) than liver patients (93% vs. 89%, +4%). The dosimetric benefit correlated with respiratory amplitude, with larger motion yielding greater improvements. The motion model demonstrated high accuracy, with a mean error of 0.1 ± 2.9 mm. Significance: This study demonstrates the feasibility and dosimetric advantage of deformable dose-guided MLC tracking using motion-model-derived volumetric imaging in patient data. These findings represent a critical step toward the clinical implementation of real-time dose-guided radiation therapy.

Objective. Anatomy continuously deforms during radiation therapy. Although real-time volumetric imaging approaches are emerging, there is a lack of adaptive strategies that account for intrafraction deformations. The purpose of this study was to develop a multileaf collimator (MLC) tracking method that adapts to deformations and evaluate the performance for lung cancer with multiple lesions. Approach. Dose-optimised deformable MLC tracking was developed using a fast dose calculation to accumulate dose at each timestep. The accumulated planned doses were deformed to represent the desired dose distribution for the deformed anatomy and the MLC leaf positions were optimised to minimise the difference between the delivered and deformed planned dose. Dose-optimised deformable MLC tracking was evaluated using four lung cancer cases generated using the 4D XCAT digital phantom. Stereotactic ablative radiotherapy treatment plans were created using a planning target volume (PTV) margin expansion of 5 mm on the gross tumour volumes (GTV). Treatments were simulated using three patient-measured motions for each phantom. The doses accumulated using the fast dose calculation model with MLC tracking were compared to an internal target volume (ITV)-based approach. Main results. The volume of the PTVs were reduced by an average of 34% using dose-optimised deformable MLC tracking compared to the ITV-based approach. The mean differences and standard deviations from the planned doses were −0.5% ± 0.6% for the GTV D _100% and −1.1% ± 0.6% for the PTV D _98% when dose-optimised deformable MLC tracking was used, and −5.2% ± 8.8% for the ITV D _100% and −13.8% ± 12.9% for the PTV D _98% when no tracking was used. Significance. The study demonstrated a proof of concept for dose-optimised deformable MLC tracking to reduce dosimetric errors for deforming anatomy. The proposed method could enable the safe reduction of treatment margins for multiple independently moving targets in the lung compared to the standard of care.

Objective: Head and neck (H&N) volumetric imaging using a standard linear accelerator during treatment delivery could enable motion management strategies to improve geometric accuracy and avoid critical organs at risk (OARs). We developed a Kalman filter (KF) tracking approach to image 3D H&N anatomy in real-time using 2D x-ray images. Approach: Our method produces 3D images and associated contours from each 2D x-ray image. The 3D images are generated by registering each 2D image to a 2D image generated using the planning CT and modelling the underlying 3D deformation vector field (DVF) within a KF to produce an updated 3D DVF. We performed a simulation study using the planning CT image and 5-minute surface motion traces for 12 H&N cancer patients. 2D intra-treatment images served as input to obtain output 3D images. We evaluated the KF tracking performance on the target and critical structures using centroid error, Hausdorff distance and Dice coefficient metrics. The results were compared to those from the current standard of care, no tracking. Results: Using KF tracking, we obtained a mean tumour centroid error of -0.2±0.6mm, 0.3±0.8mm, and 0.1±0.4.mm in the left-right, anterior-posterior and superior-inferior axes respectively, compared to 0.9±1.0mm, -0.8±1.1mm, and -0.1±1.2mm with no tracking. The 95th percentile Hausdorff distance for the tumour contour was 1.2±0.6 mm with KF tracking and 2.0±1.2 mm without tracking. The Dice coefficients were 0.91±0.08, 0.93±0.04, 0.93±0.05, 0.79±0.18, and 0.89±0.06 for the tumour, brainstem, left parotid gland, optic chiasm, and spinal cord respectively with KF tracking and 0.78±0.17, 0.84±0.06, 0.83±0.08, 0.39±0.14, and 0.70±0.13 with no tracking. Significance: Volumetric imaging with a standard linear accelerator was investigated using KF tracking with H&N patient data. The findings demonstrate accurate tracking of both the tumour and OARs, indicating potential applicability for the proposed method in a clinical translation setting.

Background: This observational study aims to establish the feasibility of using x-ray images of radio-opaque chemoembolisation deposits in patients as a method for real-time image-guided radiation therapy of hepatocellular carcinoma. Methods: This study will recruit 50 hepatocellular carcinoma patients who have had or will have stereotactic ablative radiation therapy and have had transarterial chemoembolisation with a radio-opaque agent. X-ray and computed tomography images of the patients will be analysed retrospectively. Additionally, a deep learning method for real-time motion tracking will be developed. We hypothesise that: (i) deep learning software can be developed that will successfully track the contrast agent mass on two thirds of cone beam computed tomography (CBCT) projection and intra-treatment images (ii), the mean and standard deviation (mm) difference in the location of the mass between ground truth and deep learning detection are ≤ 2 mm and ≤ 3 mm respectively and (iii) statistical modelling of study data will predict tracking success in 85% of trial participants. Discussion: Developing a real-time tracking method will enable increased targeting accuracy, without the need for additional invasive procedures to implant fiducial markers. Trial registration: Registered to ClinicalTrials.gov (NCT05169177) 12th October 2021.