Extending longitudinal field-of-view for cone-beam CT based on a novel surface-aware diffusion model

Objective.The limited longitudinal field-of-view (FOV) of cone-beam computed tomography (CBCT) restricts patient setup and adaptive radiotherapy (ART). This study aims to restore the longitudinal CBCT anatomy by introducing updated optical surface image (OSI) guidance into a diffusion-based FOV extension framework.Approach.Longitudinal CBCT FOV extension was formulated as a surface-aware image restoration task. A denoising diffusion implicit model (DDIM)-based RePaint was developed to use the truncated CBCT on the treatment-day, prior planning computed tomography (pCT), and updated OSI as complementary inputs, where pCT provided patient-specific anatomical prior, while OSI supplied synchronized surface information with CBCT to constrain inter-fractional differences from pCT. Various noise levels were simulated to test the stability. RePaint was used to improve consistency between known and restored regions, and DDIM sampling was adopted to accelerate three-dimensional inference. To test its performance, three CBCTs on different days and one pCT of 41 thoracic patients from two centers were retrospectively selected (Center A: 34 training, 4 testing, Trilogy; Center B: 3 testing, Ethos), yielding 102 training pairs and 21 test pairs.Main results.On testing cases, DDIM-RePaint achieved MAE = 25.282.45 HU, SSIM = 94.890.73, PSNR = 23.050.83 for center A, and MAE = 35.204.30 HU, SSIM = 91.272.50, PSNR = 21.120.89 for center B, both were significantly better than that of fusing pCT and CBCT as the current clinical approach (all P0.001). The inference time on 12 samples was reduced from 45 min - 2.2 min by sampler optimization.Significance.The proposed surface-aware DDIM-RePaint framework provided an accurate and efficient approach for longitudinal CBCT FOV extension, potentially facilitating patient setup and ART planning without extra imaging dose and hardware upgrade.