Projection-domain metal artifact correction using a dual layer detector

Abstract

Metal artifact remains a challenge in cone-beam CT images. Many two-pass metal artifact reduction methods have been proposed, which work fairly well, but are limited when the metal is outside the scan field-of-view (FOV) or when the metal is moving during the scan. In the former, even reconstructing with a larger FOV does not guarantee a good estimate of metal location in the projections; and in the latter, the metal location in each projection is difficult to identify due to motion. Furthermore, two-pass methods increase the total reconstruction time. In this study, a projection-based metal detection and correction method with a dual layer detector is investigated. The dual layer detector provides dual energy images with perfect temporal and spatial registration in each projection, which aid in the identification of metal. A simple phantom with metal wires (copper) and a needle (steel) is used to evaluate the projection-based metal artifact reduction method from a dual layer scan and compared with that of a single layer scan. Preliminary results showed enhanced ability to identify metal regions, leading to substantially reduced metal artifact in reconstructed images. In summary, an effective single-pass, projection-domain method using a dual layer detector has been demonstrated, and it is expected to be robust against truncation and motion.

1. PURPOSE & INTRODUCTION

Cone-beam CT is increasingly being used in numerous applications, including image-guided interventions, image-guided radiation therapy, and dedicated head, dental, and extremity scanners. In each of these applications, metal is commonplace, including devices, wires, implants, dental fillings, etc. Many metal artifact reduction (MAR) methods have been developed (see topical review [1]), most of which use a two-pass reconstruction to identify metal regions in projections (Fig. 1). Yet, there remain challenging cases for such methods, including when the metal is outside the scan FOV (Fig. 2), at interfaces where it crosses in and out of the FOV, and when the metal is moving [2]. These could largely be addressed if metal could be detected in each projection (Fig. 3). However, segmenting metal in projections is not a trivial task due to complex background anatomy and wide variations in the shape and size of metal.

Figure 1.

Conventional two-pass metal correction with untruncated projections. All metal is detected in the first-pass reconstruction and corrected in the sinogram. The resulting second-pass reconstruction is free of metal artifact. Optionally, the segmented metal can be added back into the image.

Figure 2.

Conventional two-pass metal correction with truncated projections. Only metal in the scan FOV is detected in the first-pass reconstruction and corrected in the sinogram. The resulting second-pass reconstruction contains substantial metal artifact from outside the field of view.

Figure 3.

Projection-based correction of truncated projections. Metal is directly detected in the projections, including metal outside of the scan FOV that passes across the object. The resulting reconstruction is free of metal artifact. Optionally, the detected metal can be reconstructed and added back into the image.

Dual energy imaging can generate material-specific images in the projection domain, including metal. Particularly, dual energy imaging with a dual layer detector is well suited to this task due to the perfect spatial and temporal registration of the dual layer detector. Other dual energy approaches such as kVp switching or dual source would struggle to provide projection-based metal detection in a CBCT acquisition due to spatial or temporal mismatch from gantry rotation, patient motion, or motion of the metal. In this work, we propose automatic detection and correction of metal in each projection using a dual layer detector.

2. METHODS

A prototype dual layer detector with active area of 43×43 cm² was used for this study (Fig. 4). The top layer has a 200 μm CsI scintillator that preferentially absorbs lower energy x-rays, while the bottom layer has 550 μm CsI that preferentially absorbs higher energy x-rays. The two layers are separated by a 1 mm copper filter that further increases spectral separation. The dual layer images were spatially registered with an affine transform that accounts for the different magnification, translation, and rotation between the two layers. The affine transform was predetermined with a one-time geometric calibration of the detector.

Figure 4.

Dual layer flat-panel detector prototype. The top layer consists of a thinner scintillator (200 um CsI), the bottom layer has a thicker scintillator (550 um CsI), and the two layers are separated by a filter (1 mm Cu).

A solid water phantom of 20 cm diameter (Multi-Energy CT Phantom, Gammex, Middleton, WI) with various soft-tissue equivalent inserts (adipose, brain, blood), including some with iodine enhancement (2–5 mg/ml I) was imaged. Two copper wires were placed inside one insert, while another two copper wires were placed 2.5 cm outside the phantom. A steel needle was also placed 2.5 cm outside the phantom. The metal placed outside the solid water phantom represents metal that might be difficult to identify in the first-pass reconstruction of two-pass MAR methods.

A CBCT scan of the phantom was acquired with 500 projections, each at 120 kVp, 10 mA, 18 ms, on a tabletop system. An additional scan without the wires or needle was also acquired for a metal-free reference. After log normalization of the projections, we compared metal detection for 1) single layer images, and 2) dual layer images. The low energy (top layer) image l_L was used as the single layer image, and a threshold on the line integrals of l_L > 4.8 was found to best separate metal from the background. The dual layer detector enabled additional selectivity for metal by suppressing other materials. In particular, we used a weighted subtraction, followed by a small threshold:

$$l_L-1.2l_H>0.1,$$

where l_H are the high energy (bottom layer) line integrals.

In both single and dual layer cases, thresholding was followed by a morphological opening operation to remove stray erroneous pixels. The segmented metal regions were then replaced by linearly interpolating between the nearest non-metal pixels. Finally, the corrected projections were reconstructed using filtered back-projection. For a consistent comparison, only the low energy images were used in all reconstructions. Thus, the high energy image was only used for metal detection in this work. In the future, the dual layer projections could be combined for improved detection efficiency and beam hardening correction. No other corrections were applied.

3. RESULTS

Sinograms of the central axial plane of the phantom are shown in Fig. 5. In the uncorrected sinogram, the traces of a pair of copper wires can be seen inside the phantom. Another pair of copper wires begins outside the phantom and passes across the phantom. The steel needle is less attenuating and also passes across the phantom. Thresholding the single layer image detects most of the inner copper wires, but only detects the outer wires as they pass across the phantom (red overlay). However, it fails to threshold the metal toward the edge and outside of the phantom due to the lower attenuation. On the other hand, the dual layer thresholding fully detects the copper wires in the sinogram and partially detects the needle.

Figure 5.

Sinograms of phantom scan with metal wires in the central axial plane. (a) Uncorrected sinogram showing metal traces inside and outside the phantom. Red overlay shows metal regions detected by (b) single layer thresholding and (c) dual layer thresholding, which are corrected by interpolation prior to reconstruction. (d) Reference sinogram without metal.

Reconstructions of the sinograms, including those corrected for metal-detected regions, are shown in Fig. 6. The uncorrected image shows strong streaks and metal artifact. The single layer image can substantially reduce metal artifact in the center of the phantom, corresponding to the correction in the center of the sinogram. However, strong metal artifact remains at the edge of the phantom. These are resolved in the dual layer projections, which selectively thresholds metal after suppressing the background. Residual metal from the needle has a minimal effect on image quality. Comparing the uncorrected and corrected (single, dual layer) images with the reference scan, we found that the root mean squared difference (RMSD) within the phantom was 548.0, 186.2, and 39.2 HU, respectively. The RMSD in the uncorrected and single layer corrected images is dominated by artifact, while that of the dual layer corrected image is limited by noise in the difference image.

Figure 6.

Reconstructed images (a) without MAR (uncorrected), (b) corrected using single layer thresholding, (c) corrected using dual layer thresholding, and (d) reference image without metal. Window: [−500, 500] HU.

4. DISCUSSION AND CONCLUSIONS

Metal artifact reduction methods still struggle when metal is outside the FOV or when motion is present. These limitations can be overcome by using a dual layer detector and a projection-based metal correction method. We have demonstrated this new method for correcting metal artifact, which is robust against truncation and motion. The method leverages the ability of dual layer detectors to selectively identify regions of metal in each projection. Furthermore, it only requires a single reconstruction step. If metal can be detected and corrected in real time, a filtered backprojection reconstruction can also be done in real time as the projections are acquired so that the reconstruction is complete as soon as the last projection is acquired.

Reconstructions of the corrected sinograms showed vast reduction in metal artifact, approaching the image quality of a metal-free scan or of a two-pass correction with untruncated projections. In this preliminary work, the bottom layer was only used to help detect metal. In the future, we will combine the layers when reconstructing the corrected image. While simple interpolation over the metal regions can greatly reduce metal artifact in the reconstructed images, we are also investigating more advanced inpainting methods beyond linear interpolation. In ongoing work, we are extending the method to more complex, anthropomorphic phantoms with relevant metal hardware, such as dental fillings in a head phantom, biopsy needles in a liver phantom, etc. An example image is shown in Fig. 7.

Figure 7.

(a) Projection of an anthropomorphic chest phantom, with a metal spring from an IV line on the phantom exterior. (b) The dual layer detector readily identifies the metal.