Catharine H Clark | Searcharxiv

arXiv2026

Image-Conditioned Diffusion Models for Quality Assurance of Organ-at-Risk Segmentations in Radiotherapy

Accurate organ-at-risk segmentation is essential for radiotherapy planning, but reviewing segmentations is time-consuming and subjective. We investigate normative modelling for segmentation error detection in head-and-neck CT, comparing a VAE framework with an image-conditioned segmentation diffusion model. Models were evaluated on RADCURE brainstem and spinal cord segmentations using simulated boundary and width perturbations. Error detection was assessed using the Dice similarity coefficient and the Distance to Agreement (DTA) between the input and reconstructed segmentations. While both models detected some simulated errors, regional DTA showed that the diffusion model localised subtle boundary errors more consistently. These results support image-conditioned diffusion reconstruction as a promising framework for localised, anatomy-aware segmentation QA.

cs.CV↗

arXiv2023

Improving the modeling of the Agility multi-leaf collimator

Robust fine tuning of multi-leaf collimator (MLC) Treatment Planning System (TPS) modeling parameters is crucial for creating an optimal beam model, particularly with the ever-increasing accuracy required for advancing techniques. Challenges arise from balancing the trade-off between multiple parameters and therefore the quality of tuning depends on the experience of the physicist and the procedures used. This is in part due to limitations of the MLC modeling within the TPS. As a result, the actual values used have been shown to vary widely between centers. We present and evaluate two different MLC transmission maps to improve the modeling of the Elekta Agility MLC in the RayStation TPS. The model prototypes were developed with discrete and continuous MLC transmission maps assigned to the tongue-and-groove and leaf tip regions. The prototypes aimed to replicate the average doses for synchronous and asynchronous sweeping gap fields, measured using a Farmer chamber in a solid water phantom. This study investigated the impact of these improvements using test fields and a wide variety of measured clinical plans in three different centers. The models achieved good accuracy and the improvements facilitated the standardization of the configuration and commissioning processes and extended the range of validity of TPS dose calculations in wide variety of treatment plans and measuring systems. The simpler MLC prototype with discrete transmission maps performed similarly to the more sophisticated one both in tests and clinical plans and constitutes a good option for routine clinical practice. The need for trade-offs were reduced and the models were successfully configured using a common set of parameters across the three centers, which is useful for reducing the workload and the risks associated with the configuration process, thus improving the accuracy and safety of radiotherapy treatments.

physics.med-ph↗