1425

Optimization of Reconstruction Parameters of Compressed Sensing STIR SEMAC for Metal Artifact Reduction MRI of Hip, Knee and Ankle Arthroplasty Implants: How many Iterations and how much Regularization is needed?
Gaurav Kumar Thawait1, Dharmdev H Joshi1, Esther Raithel2, Mathias Nittka2, Wesley D Gilson3, and Jan Fritz1

1Johns Hopkins University, Baltimore, MD, United States, 2Siemens Healthcare GmbH, 3Siemens Healthcare USA

Synopsis

Compressed sensing-(CS)-based Slice Encoding for Metal Artifact Correction (SEMAC) turbo spin echo (TSE) pulse sequences achieve high–quality metal artifact reduction MRI around arthroplasty implants. Compressed sensing-based undersampling of k-space permits the time-neutral use of SEMAC, but requires iterative reconstruction algorithms, which are time consuming. We determined minimum number of iterations and regularization required for diagnostic image quality of STIR CS-SEMAC data sets of hip, knee and ankle arthroplasty implants. Our results show that 15-17 iterations and 0.0035 regularization results in optimal image quality of STIR CS-SEMAC images, which currently requires 4-5 minutes of reconstruction time.

Introduction

Compressed sensing-(CS)-based acceleration for Slice Encoding for Metal Artifact Correction (SEMAC) turbo spin echo (TSE) pulse sequence has been shown to be feasible for high–quality metal artifact reduction MRI around arthroplasty implants [1-2]. When compared to 3-fold accelerated conventional SEMAC TSE, an 8-fold-accelerated CS-SEMAC pulse sequence can shorten the sampling time by up to 60% and reach acquisition times that are similar to optimized, high-bandwidth TSE pulse sequences, thus enabling the time neutral use of SEMAC [3]. CS-SEMAC reconstructions can be performed inline; however, current reconstruction times for CS-SEMAC data sets vary between 3-6 min depending on acquisition parameters such as number of slices and spatial resolution, as well as reconstruction parameters such as the number of iterations and regularization. A higher number of iterations improves image clarity and higher regularization decreases image noise, but can be time consuming. Thus, CS-SEMAC reconstructions should be performed with the minimum number of iterations and regularization that is required to achieve diagnostic image quality.

Purpose

To optimize the length of time needed for image reconstruction of STIR CS-SEMAC data sets of hip, knee and ankle arthroplasty implants through the identification of the optimal minimum of the number of iterations and regularization that is required for diagnostic image quality.

Methods

This study was approved by our institutional review board. Informed consent was obtained from all participating subjects. Three subjects underwent 1.5T MR imaging (MAGNETOM Aera, Siemens Healthcare, Erlangen, Germany) of the hip, knee and ankle, using a 15-channel transmit-receive knee coil, 16-channel receive-only ankle coil and 18-channel receive-only surface and embedded spine array coils for hips. The study protocol consisted of 8-fold accelerated STIR CS-SEMAC pulse sequences (Figure 1), which used 19 SEMAC-encoding steps and acceleration through incoherent undersampling of the 2D-phase encoding matrix and non-linear, SENSE-type reconstruction with L1-norm-based regularization was used [4]. CS-SEMAC raw data of hip, knee and ankle arthroplasty implants were reconstructed using 1, 3, 5, 7, 10, 13, 15, 17, 20, 30, 40 and 50 iterations and regularization values of 0.0005. 0.0015, 0.0025, 0.0035, 0.0045 and 0.0055. To determine the optimal minimum number of iterations that is required to achieve diagnostic image quality, all possible combinations of iterations and regularization were created. All reconstructed datasets were reviewed in consensus by two observers for their diagnostic image quality taking into consideration the amount of implant-induced susceptibility artifact as well as visibility of the implant-bone interface and periprosthetic soft tissues. The reconstruction time for all image reconstructions was recorded as well.

Results

A total of 216 reconstructions of CS-SEAMC data of the hip, knee and ankle were successfully created and reviewed. Across all iterations, the image quality achieved with a regularization of 0.0035 was ranked highest in 83.3% (10/12) of hip data sets, 100% (12/12) of knee data sets and 83.3% (10/12) of ankle data sets, whereas higher regularizations did not result in improving image quality, but resulted in edge blurring at regularization in excess of 0.0035 (Figure 2). Across all regularization parameters, the image quality achieved with 17 iterations was ranked highest in 66.7% (4/6) of hip data sets and the image quality achieved with 15 iterations was ranked highest in both 100% (6/6) of knee and 100% (6/6) of ankle data sets, whereas higher numbers of iterations did not result in improving image quality (Figure 3). Image reconstruction times increased with increasing iterations ranging from <1 minutes for 1 iteration to 25 minutes for 50 iterations (Figure 4).

Discussion

The image quality of STIR CS-SEMAC data that were acquired through pseudo-random undersampling, require iterative reconstruction that is dependent on the number of iterations and degree of regularization. While the reconstruction time is mainly a function of the number of iteration steps, our study shows that beyond certain threshold numbers of iterations and regularization, no additional gain in image quality may be achieved. The thresholds are 17 iterations for hip and 15 iterations for knee and ankle CS-SEMAC STIR data sets, which require reconstruction times of 5 minutes, 4 minutes and 4 minutes respectively. This has important practical implications as reconstruction times are substantial for datasets of more than 20 iterations. Similarly, the ideal regularization was found to be 0.0035 for STIR CS-SEMAC data sets. Increasing the regularization beyond this threshold did not result in improving image quality, but introduce blurring of contour with higher regularization.

Conclusion

Optimal image quality of STIR CS-SEMAC MR images of hip, knee and ankle may be achieved with regularization parameter of 0.0035 and minimum of 15-17 iterations, which has important practical implication to limit the required reconstruction time.

Acknowledgements

No acknowledgement found.

References

1. Lu W, Pauly KB, Gold GE, Pauly JM, Hargreaves BA. SEMAC: Slice Encoding for Metal Artifact Correction in MRI. Magn Reson Med. 2009 Jul;62(1):66-76. doi: 10.1002/mrm.21967.

2. Fritz J, Ahlawat S, Demehri S, Thawait GK, Raithel E, Gilson WD, Nittka M. Compressed Sensing SEMAC: 8-fold Accelerated High Resolution Metal Artifact Reduction MRI of Cobalt-Chromium Knee Arthroplasty Implants. Invest Radiol. 2016 Oct;51(10):666-76

3. Fritz J, Fritz B, Thawait GK, Raithel E, Gilson WD, Nittka M, Mont MA. Advanced metal artifact reduction MRI of metal-on-metal hip resurfacing arthroplasty implants: compressed sensing acceleration enables the time-neutral use of SEMAC. Skeletal Radiol. 2016 Oct;45(10):1345-56.

4. Liu j, Rapin J, Chang T, Schmit P, Bi X, Lefebvre A, Zenge M, Mueller E, Nadar M. Regularized reconstruction using redundant Haar wavelets: A means to achieve high under-sampling factors in non-contrast-enhanced 4D MRA. ", Proceeding of the International Society for Magnetic Resonance in Medicine, 20TH Annual Meeting and Exhibition, Melbourne, Australia, 5-11 May 2012, Vol. 20, 21 April 2012 (2012-04-21), pages 2237.

Figures

Figure 1. Compressed-sensing SEMAC MRI study protocol.

Figure 2. Coronal CS-SEMAC STIR MR images of the hip reconstructed with 0.0035 regularization and different number of iterations, which are stated in the right lower corner of the images.

Figure 3. Coronal CS-SEMAC STIR MR images of the hip reconstructed with 17 iterations and different regularization parameters, which are stated in the right lower corner of the images.

Figure 4. Graphs show the reconstruction times of CS-SEMAC STIR MR images of the hip, knee and ankle in relation to regularization parameters and number of iterations.

Proc. Intl. Soc. Mag. Reson. Med. 25 (2017)
1425