Compressed-sensing multispectral imaging of the postoperative spine

Abstract

Purpose: To apply compressed sensing (CS) to in vivo multispectral imaging (MSI), which uses additional encoding to avoid magnetic resonance imaging (MRI) artifacts near metal, and demonstrate the feasibility of CS-MSI in postoperative spinal imaging.

Materials and methods: Thirteen subjects referred for spinal MRI were examined using T2-weighted MSI. A CS undersampling factor was first determined using a structural similarity index as a metric for image quality. Next, these fully sampled datasets were retrospectively undersampled using a variable-density random sampling scheme and reconstructed using an iterative soft-thresholding method. The fully and undersampled images were compared using a 5-point scale. Prospectively undersampled CS-MSI data were also acquired from two subjects to ensure that the prospective random sampling did not affect the image quality.

Results: A two-fold outer reduction factor was deemed feasible for the spinal datasets. CS-MSI images were shown to be equivalent or better than the original MSI images in all categories: nerve visualization: P = 0.00018; image artifact: P = 0.00031; image quality: P = 0.0030. No alteration of image quality and T2 contrast was observed from prospectively undersampled CS-MSI.

Conclusion: This study shows that the inherently sparse nature of MSI data allows modest undersampling followed by CS reconstruction with no loss of diagnostic quality.

Figure 1

Approximate Message Passing (AMP), a variant of iterative soft thresholding, is used in the CS reconstruction. The 2D Fourier transform (in y𒀓z) and Daubechies 4 wavelet transform were used. The soft-thresholding step is data-driven and based on the undersampling factor and signal level in the highest wavelet sub-band. The AMP term (previous iteration’s residual multiplied by AMP weight) enables fast computation. The stopping criterion is met when the normalized difference of the L2-norm of successive residuals is less than 0.1%.

Figure 2

Sagittal spine images of the fully-sampled half-Fourier and reconstructed at 2–3× outer reduction factors, with SSIM maps (bottom) obtained by comparison to the fully-sampled image. The values refer to the SSIM average ± standard deviation (minimum:maximum) in the ROI.

Figure 3

Sagittal T2-weighted FSE and MSI slices from two subjects (top and bottom row) acquired at 1.5 T demonstrating the retrospective application of compressed sensing. 2D FSE images are shown to show the distortion induced by the presence of metal. The arrows point to nerves and neural foramina that are clearly depicted in the images acquired with MSI. The percentage sampled in the fully-sampled half-Fourier image is under 50% as corners in ky-kz space were not acquired.

Figure 4

(a) Acquired k-space plots in the reference regularly sampled acquisition and the CS randomly sampled acquisition with reduction factors of 2 and 3 respectively. The k-space plots correspond to the MSI images in (b). The colors refer to the echo numbers along the echo train (echo-train length = 20). Three example echo-train pathways (black lines) are shown in each k-space. (b) Sagittal T2-weighted acquisitions at 3 T with 2D fast spin echo (FSE), reference MSI with 2× parallel imaging in the anterior-posterior direction (acquisition time = 10 min) and CS-MSI with 3× outer reduction (acquisition time = 8:13 min). The solid arrows point to a neural foramen that is obscured on the 2D FSE image due to severe signal distortion. A slight parallel imaging artifact was seen (dashed arrow) due to the use of a linear coil array with sensitivity variation largely in the superior-inferior direction. Despite the slight blurring in the CS image, the overall image quality and contrast between both MSI acquisitions are comparable.