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.

High-resolution spiral CT of the breast at very low dose: concept and feasibility considerations.

OBJECTIVE: Mammography, today's standard imaging approach, has deficits with respect to the superimposition of anatomical structures. Dedicated CT of the breast so far indicated that it can provide superior soft-tissue imaging, but that it still has significant limitations with respect to spatial resolution and dose. We have assessed novel dedicated breast CT technology. METHODS: Based on simulations and measurements we developed novel technology which uses direct-conversion CdTe material and photon-counting electronics with 100 µm detector element size for close to 100% dose efficiency. We assessed the potential for the imaging of microcalcifications of 100 to 200 µm diameter and soft-tissue lesions of 1 to 5 mm diameter by simulations at dose levels between 1 and 6 mGy. RESULTS: Microcalcifications of 150 µm and soft-tissue lesions of 2 mm diameter were found to be clearly detectable at an average glandular dose of 3 mGy. Separate displays are required for high-resolution microcalcification and for low-resolution soft-tissue analysis. Total CT data acquisition time will be below 10 s. CONCLUSION: Dedicated breast CT may eventually provide comprehensive diagnostic assessment of microcalcifications and soft-tissue structures at dose levels equivalent to or below those of two-view screening mammography.

Fast on-site Monte Carlo tool for dose calculations in CT applications.

PURPOSE: Monte Carlo (MC) simulation is an established technique for dose calculation in diagnostic radiology. The major drawback is its high computational demand, which limits the possibility of usage in real-time applications. The aim of this study was to develop fast on-site computed tomography (CT) specific MC dose calculations by using a graphics processing unit (GPU) cluster. METHODS: GPUs are powerful systems which are especially suited to problems that can be expressed as data-parallel computations. In MC simulations, each photon track is independent of the others; each launched photon can be mapped to one thread on the GPU, thousands of threads are executed in parallel in order to achieve high performance. For further acceleration, the authors considered multiple GPUs. The total computation was divided into different parts which can be calculated in parallel on multiple devices. The GPU cluster is an MC calculation server which is connected to the CT scanner and computes 3D dose distributions on-site immediately after image reconstruction. To estimate the performance gain, the authors benchmarked dose calculation times on a 2.6 GHz Intel Xeon 5430 Quad core workstation equipped with two NVIDIA GeForce GTX 285 cards. The on-site calculation concept was demonstrated for clinical and preclinical datasets on CT scanners (multislice CT, flat-detector CT, and micro-CT) with varying geometry, spectra, and filtration. To validate the GPU-based MC algorithm, the authors measured dose values on a 64-slice CT system using calibrated ionization chambers and thermoluminesence dosimeters (TLDs) which were placed inside standard cylindrical polymethyl methacrylate (PMMA) phantoms. RESULTS: The dose values and profiles obtained by GPU-based MC simulations were in the expected good agreement with computed tomography dose index (CTDI) measurements and reference TLD profiles with differences being less than 5%. For 10(9) photon histories simulated in a 256 × 256 × 12 voxel thorax dataset with voxel size of 1.36 × 1.36 × 3.00 mm(3), calculation times of about 70 and 24 min were necessary with single-core and multiple-core central processing unit (CPU) solutions, respectively. Using GPUs, the same MC calculations were performed in 1.27 min (single card) and 0.65 min (two cards) without a loss in quality. Simulations were thus speeded up by factors up to 55 and 36 compared to single-core and multiple-core CPU, respectively. The performance scaled nearly linearly with the number of GPUs. Tests confirmed that the proposed GPU-based MC tool can be easily adapted to different types of CT scanners and used as service providers for fast on-site dose calculations. CONCLUSIONS: The Monte Carlo software package provides fast on-site calculation of 3D dose distributions in the CT suite which makes it a practical tool for any type of CT-specific application.

High-speed slot-scanning radiography using small-angle tomosynthesis: Investigation of spatial resolution.

PURPOSE: This paper studies spatial resolution that is achievable with a fast slot-scanning tomosynthesis approach for orthopedic examinations. Hereby, we use parallel scanning motion implemented in a twin robotic x-ray system. METHODS: We have measured and analyzed the modulation transfer function (MTF) for various combinations of scanning speed, x-ray tube voltage, pulse length, the nominal focal spot size as well as source-to-object distances. Moreover, we present a theoretical model which describes the system in terms of the MTF. The system was equipped with newly developed linear trajectory prototypes for slot scanning. The acquired images form the basis for a small-angle tomosynthesis reconstruction. In total, three different scanning speeds (27, 14, 8 cm/s), pulse lengths (1, 2, 4 ms), tube voltages (80, 100, 120 kV), two nominal focal spot sizes (0.6, 1.0), and three source-to-object distances (950, 1050, 1150 mm) were investigated. To determine the resolution capabilities, we measured the MTF for the given parameter space. The results were then used to design a filter that yields a desired resolution in the reconstructed image. In addition, we also measured the noise power spectrum (NPS) to show the influence of the aforementioned filters on the noise distribution. RESULTS: We have shown that the presented model is in good agreement with the performed measurements. Scanning speed and pulse width have an impact on the MTF in the scanning direction. Up to a travel distance of 0.3 mm during an x-ray pulse, an isotropic resolution can be achieved. Longer pulse width or higher scanning speed cause anisotropic resolution. Moreover, it is shown that none of the investigated parameters have an influence on the MTF perpendicular to the scanning direction (slot direction). The 10% MTF value ranges between 9 and 18 lp/cm in the scanning direction and about 18 lp/cm in slot direction. Tube voltage, nominal focal spot size as well as the source-to-object distance showed no major impact on the system MTF. In terms of the anisotropic resolution capabilities, we have shown that limiting the resolution in the slot direction to obtain isotropic resolution is possible yet at the cost of an inhomogeneous noise pattern. However, maintaining the resolution in slot direction will provide a better edge response and a more homogeneous noise texture at the cost of an inhomogeneous image resolution. CONCLUSIONS: We have demonstrated the feasibility of the slot-scanning technique using a twin-robotic x-ray system. Even the fastest scanning mode (27 cm/s) yields image resolution on a level that is sufficient for typical orthopedic examinations in terms of musculoskeletal (MSK) measurements. Moreover, it could be shown that the application of specifically designed target MTFs on two-dimensional x-ray images is feasible.

Generating and using patient-specific whole-body models for organ dose estimates in CT with increased accuracy: feasibility and validation.

The estimation of patient dose using Monte Carlo (MC) simulations based on the available patient CT images is limited to the length of the scan. Software tools for dose estimation based on standard computational phantoms overcome this problem; however, they are limited with respect to taking individual patient anatomy into account. The purpose of this study was to generate whole-body patient models in order to take scattered radiation and over-scanning effects into account. Thorax examinations were performed on three physical anthropomorphic phantoms at tube voltages of 80 kV and 120 kV; absorbed dose was measured using thermoluminescence dosimeters (TLD). Whole-body voxel models were built as a combination of the acquired CT images appended by data taken from widely used anthropomorphic voxel phantoms. MC simulations were performed both for the CT image volumes alone and for the whole-body models. Measured and calculated dose distributions were compared for each TLD chip position; additionally, organ doses were determined. MC simulations based only on CT data underestimated dose by 8%-15% on average depending on patient size with highest underestimation values of 37% for the adult phantom at the caudal border of the image volume. The use of whole-body models substantially reduced these errors; measured and simulated results consistently agreed to better than 10%. This study demonstrates that combined whole-body models can provide three-dimensional dose distributions with improved accuracy. Using the presented concept should be of high interest for research studies which demand high accuracy, e.g. for dose optimization efforts.

Iterative reconstruction methods in X-ray CT.

Iterative reconstruction (IR) methods have recently re-emerged in transmission x-ray computed tomography (CT). They were successfully used in the early years of CT, but given up when the amount of measured data increased because of the higher computational demands of IR compared to analytical methods. The availability of large computational capacities in normal workstations and the ongoing efforts towards lower doses in CT have changed the situation; IR has become a hot topic for all major vendors of clinical CT systems in the past 5 years. This review strives to provide information on IR methods and aims at interested physicists and physicians already active in the field of CT. We give an overview on the terminology used and an introduction to the most important algorithmic concepts including references for further reading. As a practical example, details on a model-based iterative reconstruction algorithm implemented on a modern graphics adapter (GPU) are presented, followed by application examples for several dedicated CT scanners in order to demonstrate the performance and potential of iterative reconstruction methods. Finally, some general thoughts regarding the advantages and disadvantages of IR methods as well as open points for research in this field are discussed.

A novel forward projection-based metal artifact reduction method for flat-detector computed tomography.

Metallic implants generate streak-like artifacts in flat-detector computed tomography (FD-CT) reconstructed volumetric images. This study presents a novel method for reducing these disturbing artifacts by inserting discarded information into the original rawdata using a three-step correction procedure and working directly with each detector element. Computation times are minimized by completely implementing the correction process on graphics processing units (GPUs). First, the original volume is corrected using a three-dimensional interpolation scheme in the rawdata domain, followed by a second reconstruction. This metal artifact-reduced volume is then segmented into three materials, i.e. air, soft-tissue and bone, using a threshold-based algorithm. Subsequently, a forward projection of the obtained tissue-class model substitutes the missing or corrupted attenuation values directly for each flat detector element that contains attenuation values corresponding to metal parts, followed by a final reconstruction. Experiments using tissue-equivalent phantoms showed a significant reduction of metal artifacts (deviations of CT values after correction compared to measurements without metallic inserts reduced typically to below 20 HU, differences in image noise to below 5 HU) caused by the implants and no significant resolution losses even in areas close to the inserts. To cover a variety of different cases, cadaver measurements and clinical images in the knee, head and spine region were used to investigate the effectiveness and applicability of our method. A comparison to a three-dimensional interpolation correction showed that the new approach outperformed interpolation schemes. Correction times are minimized, and initial and corrected images are made available at almost the same time (12.7 s for the initial reconstruction, 46.2 s for the final corrected image compared to 114.1 s and 355.1 s on central processing units (CPUs)).