Robustness of deep learning-based denoising 4DCBCT methods

Abstract

Objective. 4D cone-beam computed tomography (4DCBCT) is a technique used to address respiratory motion in radiotherapy but is limited by significant view-aliasing artifacts. Recently, deep learning methods have been proposed to reduce view-aliasing. This study investigates the robustness and performance of these methods when variations in patient breathing period affect the noise pattern in the scans used as training data, which potentially affects the ability of the models to reduce view-aliasing. Approach. We evaluated both a supervised and a self-supervised scalable deep learning method. Using a dataset of 328 patients, scans were partitioned into cohorts according to breathing period, and independent training with cross-cohort validation was performed to assess robustness. We also evaluated the impact of a modern UNet-based architecture and data augmentation. Main results. Training and testing on matched breathing-period cohorts did not consistently improve performance, indicating that both methods are robust to clinically observed variation in average breathing period. Instead, increasing the overall dataset size led to the most noticeable improvements. For the cohort with long breathing periods, the supervised model trained on the combined dataset achieved a ≈0.5 PSNR improvement over the cohort-specific model and a PSG improvement of ≈0.7. Significance. These findings support the reliability of deep learning-based 4DCBCT denoising across the observed range of average breathing periods. They suggest that large, diverse datasets should be prioritized over breathing-cohort-specific optimization. Finally, PSG provides a quantitative indicator that is sensitive to view-aliasing and may support future development of more robust 4DCBCT enhancement techniques.