Simulation-driven training of vision transformers enables metal artifact reduction of highly truncated CBCT scans.

BACKGROUND: Due to the high attenuation of metals, severe artifacts occur in cone beam computed tomography (CBCT). The metal segmentation in CBCT projections usually serves as a prerequisite for metal artifact reduction (MAR) algorithms. PURPOSE: The occurrence of truncation caused by the limited detector size leads to the incomplete acquisition of metal masks from the threshold-based method in CBCT volume. Therefore, segmenting metal directly in CBCT projections is pursued in this work. METHODS: Since the generation of high quality clinical training data is a constant challenge, this study proposes to generate simulated digital radiographs (data I) based on real CT data combined with self-designed computer aided design (CAD) implants. In addition to the simulated projections generated from 3D volumes, 2D x-ray images combined with projections of implants serve as the complementary data set (data II) to improve the network performance. In this work, SwinConvUNet consisting of shift window (Swin) vision transformers (ViTs) with patch merging as encoder is proposed for metal segmentation. RESULTS: The model's performance is evaluated on accurately labeled test datasets obtained from cadaver scans as well as the unlabeled clinical projections. When trained on the data I only, the convolutional neural network (CNN) encoder-based networks UNet and TransUNet achieve only limited performance on the cadaver test data, with an average dice score of 0.821 and 0.850. After using both data II and data I during training, the average dice scores for the two models increase to 0.906 and 0.919, respectively. By replacing the CNN encoder with Swin transformer, the proposed SwinConvUNet reaches an average dice score of 0.933 for cadaver projections when only trained on the data I. Furthermore, SwinConvUNet has the largest average dice score of 0.953 for cadaver projections when trained on the combined data set. CONCLUSIONS: Our experiments quantitatively demonstrate the effectiveness of the combination of the projections simulated under two pathways for network training. Besides, the proposed SwinConvUNet trained on the simulated projections performs state-of-the-art, robust metal segmentation as demonstrated on experiments on cadaver and clinical data sets. With the accurate segmentations from the proposed model, MAR can be conducted even for highly truncated CBCT scans.

DL-based inpainting for metal artifact reduction for cone beam CT using metal path length information.

BACKGROUND: Metallic implants, which are inserted into the patient's body during trauma interventions, are the main cause of heavy artifacts in 3D X-ray acquisitions. These artifacts then hinder the evaluation of the correct implant's positioning, thus leading to a disturbed patient's healing process and increased revision rates. PURPOSE: This problem is tackled by so-called metal artifact reduction (MAR) methods. This paper examines possible advances in the inpainting process of such MAR methods to decrease disruptive artifacts while simultaneously preserving important anatomical structures adjacent to the inserted implants. METHODS: In this paper, a learning-based inpainting method for cone-beam computed tomography is proposed that couples a convolutional neural network (CNN) with an estimated metal path length as prior knowledge. Further, the proposed method is solely trained and evaluated on real measured data. RESULTS: The proposed inpainting approach shows advantages over the inpainting method used by the currently clinically approved frequency split metal artifact reduction (fsMAR) method as well as the learning-based state-of-the-art (SOTA) method PConv-Net. The major improvement of the proposed inpainting method lies in the ability to correctly preserve important anatomical structures in those regions adjacent to the metal implants. Especially these regions are highly important for a correct implant's positioning in an intraoperative setup. Using the proposed inpainting, the corresponding MAR volumes reach a mean structural similarity index measure (SSIM) score of 0.9974 and outperform the other methods by up to 6 dB on single slices regarding the peak signal-to-noise ratio (PSNR) score. Furthermore, it can be shown that the proposed method can generalize to clinical cases at hand. CONCLUSIONS: In this paper, a learning-based inpainting network is proposed that leverages prior knowledge about the metal path length of the inserted implant. Evaluations on real measured data reveal an increased overall MAR performance, especially regarding the preservation of anatomical structures adjacent to the inserted implants. Further evaluations suggest the ability of the proposed approach to generalize to clinical cases.