Image-Conditioned Diffusion Models for Quality Assurance of Organ-at-Risk Segmentations in Radiotherapy
2026-08-24 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors studied ways to find mistakes in organ outlines used for radiation treatment planning in head-and-neck CT scans. They compared two computer methods: one using a variational autoencoder (VAE) and another using a segmentation diffusion model that considers the image details. They tested how well these methods detected intentionally made small errors in brainstem and spinal cord outlines. The authors found that both methods could spot some errors, but the diffusion model was better at pinpointing subtle boundary mistakes. This suggests that image-conditioned diffusion models might help improve the checking of organ segmentations in medical images.
organ-at-risk segmentationradiotherapy planninghead-and-neck CTvariational autoencoder (VAE)segmentation diffusion modelDice similarity coefficientDistance to Agreement (DTA)brainstem segmentationspinal cord segmentationsegmentation error detection
Authors
Clea Dronne, Catharine H Clark, Xavier Loizeau, Elizabeth Miles, Peter Hoskin, Jamie R McClelland
Abstract
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.