Improved field-mapping and artifact correction in multispectral imaging

Abstract

Purpose: To develop a method for improved B₀ field-map estimation, deblurring, and image combination for multispectral imaging near metal.

Methods: A goodness-of-fit field-map estimation technique is proposed that uses only the multispectral imaging (MSI) data to estimate the field map. Using the improved field map, a novel deblurring technique is proposed that also employs a new image combination scheme to reduce the effects of noise and other residual MSI artifacts. The proposed field-map estimation and deblurring techniques are compared to the current methods in phantoms and/or in vivo from subjects with knee, hip, and spinal metallic implants.

Results: Phantom experiments validate that the goodness-of-fit field-map estimation is less sensitive to noise and bias than the conventional center-of-mass technique, which reduces distortion in the deblurring methods. The new deblurring approach also is substantially less sensitive to noise and distortion than the current deblurring method, as demonstrated in phantoms and in vivo, and is able to find a good tradeoff between deblurring and distortion.

Conclusion: The proposed methods not only enable field-mapping with reduced noise sensitivity but are able to create deblurred images with less distortion and better signal-to-noise ratio with no additional scan time, thereby enabling improved visualization of underlying anatomy near metallic implants. Magn Reson Med 78:2022-2034, 2017. © 2017 International Society for Magnetic Resonance in Medicine.

Keywords: MRI of arthroplasty; field-mapping; image artifacts; metal implants.

FIG. 1

MSI images of a digital phantom with linear off-resonance in the phase-encode direction (a–e) and agar phantom (f–o) are shown demonstrating the tradeoff between blurring and distortion. The left three columns are the central MAVRIC-SL spectral bin images, each demodulated at the bin center frequency to remove large bulk distortion. The RSOS images of all bins demodulated at the bin center frequency, which blurs the images but removes bulk distortion (solid arrows), are shown in the fourth column. Conversely, the RSOS of all bin images demodulated at the same center frequency are shown in the rightmost column and exhibit deblurring with distortion in regions of off-resonance (dotted arrows). The goal of deblurring methods in MSI is to find a method that creates deblurred images without introducing distortion.

FIG. 2

An ideal bin signal profile across spectral bins is shown (top) along with the same signal with added noise (bottom) and the RF profile centered on the correct off-resonance frequency (black line in both top and bottom). Both the center-of-mass and matched-filter field-map estimates are accurate in the absence of noise. However, in the presence of noise, the center-of-mass estimate becomes noticeably biased towards the frequency of the center bin, while the matched-filter estimator remains unbiased by still accurately estimating the correct off-resonance frequency.

FIG. 3

The proposed weighted-average deblurring reconstruction algorithm is outlined. Bin images are first used to create the goodness-of-fit field-map estimate and the goodness-of-fit metric, which predicts the accuracy of the field-map estimate based on how closely the actual and expected bin signal profiles match. Using the bin images and field-map estimate, a deblurred goodness-of-fit reconstructed image is then created using the expected RF-weighted combination across bins. The weighted-average reconstruction then uses the goodness-of-fit to reduce the effects of errors in the field-map estimate, thus striking a good balance between the blurring seen in the RSOS image and distortion seen in the goodness-of-fit reconstruction when the field-map estimate is not correct.

FIG. 4

A field-map comparison from a digital metallic implant phantom is shown between the center-of-mass and goodness-of-fit field-map estimations. Solid arrows indicate that the goodness-of-fit method is far less sensitive to noise added to the digital phantom than the center-of-mass method. The center-of-mass method is biased towards the center bin (i.e., 0 Hz) (dotted arrows), whereas bias in the goodness-of-fit method is far less pronounced and based on the direction and magnitude of the off-resonance gradient (dashed arrows).

FIG. 5

Images are shown from a digital metallic implant phantom processed using the RSOS bin combination (a,b,d) and the RF-weighted bin combination (c,e). Both the center-of-mass field-map (b,c) and goodness-of-fit field-map (d,e) estimates were used for deblurring. The final weighted-average reconstruction is also shown (f). The RSOS bin combined image (a) has no distortion but has blurring throughout the entire image. All deblurring methods deblur well in small off-resonance regions away from the metal (solid arrows). However, those that use the RSOS bin combination (b,d) can include energy from distant spectral frequencies (dashed arrows). Bias in the center-of-mass field-map estimate can create distortion during deblurring (dotted arrows). When both of these artifacts are combined (b), the overall image, although deblurred away from the metal, can appear worse than the blurry, undistorted version. The proposed weighted-average reconstruction, on the other hand, performs well across the entire image and has substantially less distortion/artifacts than the current center-of-mass technique (b), while still providing good deblurring. Although some pixilation artifacts in the simulated data are unrealistic, the pile-up, ripple and blurring artifacts are very similar to what is seen in phantom and human scans.

FIG. 6

The RSOS (left column), center-of-mass (middle column), and weighted-average (right column) reconstructions of an agar grid phantom with a metallic hip implant. Zoomed in portions of a high SNR region are shown (d–f) from the same slice as the top row. A zoomed in low SNR region from an edge slice that has a large off-resonance due to the VAT gradient is also shown (g–i). In regions of high SNR and low field inhomogeneity both deblurring methods perform well (solid arrows). In regions of large off-resonance the weighted-average method removes the distortion/displacement seen in the center-of-mass method due to the decreased bias in the goodness-of-fit field-map estimate (dotted arrows). For low SNR regions, the weighted-average method led to improved SNR and deblurring compared to the center-of-mass method (dashed arrows).

FIG. 7

Noise analysis of an agar grid phantom with a metallic implant highlights the noise performance of the RSOS, center-of-mass, and weighted-average reconstructions. Relative magnitude mean errors (f–h) and SNR maps (k–m) show that the weighted-average method is less sensitive to noise related bias and variance than the center-of-mass method, which generally has worse performance than the RSOS technique. Additionally, noise analysis was done to compare the noise characteristics of the center-of-mass and goodness-of-fit field-map estimates. The larger range of estimated frequencies in the goodness-of-fit field-map indicates that it does not suffer from the same bias as the center-of-mass method. The magnitude mean error (i–j) and standard deviation (n–o) images show that the goodness-of-fit field-map is substantially less biased and noisy than the center-of-mass field-map estimate.

FIG. 8

RSOS, center-of-mass, and weighted-average reconstructions of MAVRIC-SL data along with the calculated goodness-of-fit metric are shown for subjects with knee (a–d), hip (e–h), and spinal (i–l) implants. Solid arrows indicate regions where both the center-of-mass and the weighted-average reconstructions deblur the image well. The weighted-average reconstruction is able to perform deblurring without introducing the severe distortion seen in the center-of-mass method (dotted arrows). The goodness-of-fit images also show that even images with a relatively low goodness-of-fit (h) are still able to produce weighted-average reconstructions (g) that deblur well across the entire image without introducing any noticeable distortion.

FIG. 9

RSOS, center-of-mass, and weighted-average reconstructions of two separate subjects with metallic hip implants are shown to visualize artifacts created by both deblurring methods. Solid red arrows point to regions where both deblurring methods produced adequate deblurring. Dotted red arrows point to regions where the proposed weighted-average reconstruction had good deblurring while reducing distortions. Dashed blue arrows point to three types of artifacts seen occasionally by the weighted-average images, including: 1) a single voxel with incorrect field-map estimation due to uncertainty in a quickly varying magnetic field (c). 2) Slight hypointense patterns when the estimated field-map has a sharp discontinuity (upper dashed blue arrow in f), this occurred because the acquired spectral profile was bimodal. 3) Slight blocky appearance at large signal boundaries due to sharp discontinuity in estimated field-map (lower blue dashed arrows in f). While these are among the worst of the artifacts associated with the weighted-average method, the images still exhibit excellent deblurring with considerably less distortion than the center-of-mass method.