Abstract
Metal artifact correction is a challenging problem in cone beam computed tomography (CBCT) scanning. Metal implants inserted into the anatomy cause severe artifacts in reconstructed images. Widely used inpainting-based metal artifact reduction (MAR) methods require segmentation of metal traces in the projections as a first step, which is a challenging task. One approach is to use a deep learning method to segment metals in the projections. However, the success of deep learning methods is limited by the availability of realistic training data. It is laborious and time consuming to get reliable ground truth annotations due to unclear implant boundaries and large numbers of projections. We propose to use X-ray simulations to generate synthetic metal segmentation training dataset from clinical CBCT scans. We compare the effect of simulations with different numbers of photons and also compare several training strategies to augment the available data. We compare our model's performance on real clinical scans with conventional region growing threshold-based MAR, moving metal artifact reduction method, and a recent deep learning method. We show that simulations with relatively small number of photons are suitable for the metal segmentation task and that training the deep learning model with full size and cropped projections together improves the robustness of the model. We show substantial improvement in the image quality affected by severe motion, voxel size under-sampling, and out-of-FOV metals. Our method can be easily integrated into the existing projection-based MAR pipeline to get improved image quality. This method can provide a novel paradigm to accurately segment metals in CBCT projections.
Introduction
Metallic implants, such as dental fillings, screws, or joint replacements, introduce severe distortions in Cone Beam Computed Tomography (CBCT) scans. These "metal artifacts" manifest as bright streaks and dark shadows that obscure critical anatomical details, complicating clinical interpretation. Standard Inpainting-based Metal Artifact Reduction (MAR) requires identifying the exact location of metal in every 2D X-ray projection—the "metal trace." However, manual labeling is impractical due to the high volume of projections, and traditional geometric calculations often fail when implants move or sit outside the scanner's standard field of view (FOV).
Methodology
To overcome the scarcity of labeled clinical data, we developed a pipeline to generate high-fidelity synthetic training sets using physics-based simulations. By taking existing clinical scans and performing Monte Carlo simulations, we projected segmented 3D metal back into 2D space. This allowed for the creation of perfectly labeled pairs of metal-corrupted projections and their corresponding ground truth masks.
The study utilized 26 clinical scans covering various anatomies (knee, wrist, ankle, etc.). We trained a modified U-Net architecture using a mix of full-sized projections and smaller image crops to prevent the model from generating false positives. We specifically investigated the impact of simulated photon counts on model robustness.
Key Insights and Results
A significant finding of this research is that "noisy" simulations (using 300–1400 photons per pixel) outperformed "clean" simulations (5000 photons). The noise inherent in low-photon simulations acts as a natural regularizer, making the model more robust to the imperfections found in real-world clinical hardware.
The proposed model, Noisy (full+crops), achieved a Mean Dice Similarity Coefficient (DSC) of 94.8%, significantly outperforming conventional methods like CMAR (90.5%) and MMAR (89.1%). Furthermore, because the model operates on 2D projections independently, it remains robust in scenarios that typically baffle 3D-based methods, such as patient motion and out-of-FOV metal placement.
Conclusion
This deep learning approach, powered by synthetic data generation, provides a robust solution for metal segmentation in CBCT. By leveraging noisy simulations, the model generalizes effectively to real clinical data. Future work will focus on expanding this methodology to dental scans and integrating the segmentation process into dual-domain restoration networks for simultaneous 2D and 3D artifact correction.
Summary updated