PROPELLER consists of rotatory blades and is well-known for motion robustness. Previous studies have investigated the feasibility of SMS PROPELLER by applying SMS in each blade [1, 2]. For reconstruction, SMS slice-separation can be done within each blade separately, which we denote as simple single-blade (SSB) approach. However, SSB approach has two limitations: excessive noise may arise from individual blades; artifact may arise when blades are narrow. Here, we propose multi-step joint-blade (MJB) SENSE reconstruction, which jointly reconstructs all blades in three steps. Compared to SSB SENSE, MJB SENSE substantially reduces noise amplification and the narrow-blade artifact so that high MB factors is feasible.

Simple Single-Blade SENSE: The slices are reconstructed using SENSE separately within each blade [2, 3], and motion correction [4] is performed for each slice independently. The blades are combined with k-space density filtering.

Multi-Step Joint-Blade SENSE: MJB SENSE consists of three steps as follows (Figure 1).

(1) SSB SENSE is performed. It provides rough estimations of blade images, blade-combined images, and motion parameters.

(2) Regularized single-blade SENSE is performed. It improves the estimations provided by SSB SENSE. From the blade-combined images, we regenerate two sets of blade images, with and without coil sensitivity map (CSM) weighting. Dividing the CSM-weighted regenerated blade images by non-CSM-weighted ones, we obtain so-called narrow-blade CSMs. By “back-substitution”, the pixels are re-estimated one-by-one in SENSE equations: when estimating one pixel, all other pixels are treated known and replaced by the corresponding pixels in the regenerated blade images. Motion parameters and blade-combined images are updated.

(3) Joint-blade SENSE is performed. It produces the final images. High frequency data is regenerated to widen the acquired blades to full resolution. Joint-blade SENSE equation is formed pixel-wise: for each slice to solve, the images of widened blades are rotated and shifted according to motion parameters and CAIPI shift patterns, such that this slice is spatially fixed in all blades; for each pixel on this slice, the single-blade SENSE equations are combined to form the joint-blade SENSE equation. Back-substitution is also used.

Simulations: SMS PROPELLER data (16 blades×30 lines or 32 blades×15 lines) were synthesized from phantom images and brain images, at MB4/6/8. The images were acquired at 3T using eight-channel coil with matrix size 256×256. Inter-blade in-plane rigid motion was added with uniform random distribution (±5 degree rotation and ±5 pixels translation). Random Gaussian noise was added with SNR=10/30. G-factor maps were generated by pseudo multiple replica [5].

(1) In phantom simulation (Figure 2), MJB SENSE resulted in much less noise than SSB SENSE. The nRMS error and mean g-factor confirmed the noise reduction using MJB SENSE was 30%-50% (Figure 3). The mean g-factors using SSB/MJB SENSE were 1.27/0.83, 3.66/1.85, and 7.46/3.81 at MB4, MB6, and MB8, respectively.

(2) Even without noise (Figure 4), SSB SENSE resulted in ringing artifact and slice leakage, which became more serious with narrower blades; MJB SENSE largely reduced these artifacts.

(3) In moving brain simulation (Figure 5), the noise reduction effect of MJB SENSE was similar to that in the phantom simulation. Both methods achieved good motion correction.

(1) MJB SENSE substantially reduces noise amplification, because it treats the data as a whole rather than separately. Specifically, we establish joint-blade SENSE equations such that the blades are linked and regularized mutually; we also utilize blade-combined images, which is essentially utilizing all blades. In fact, similar concept has also been employed in other reconstruction methods to improve SNR [6,7]. With the capability of reducing noise, MJB SENSE may enable very high MB factors for future SMS PROPELLER applications.

(2) MJB SENSE largely reduces the narrow-blade artifact. Normal Cartesian SENSE assumes point-spread-function being Dirac [8], which is not satisfied with narrow blades. To address this, customized CSMs are used for the blades in step 2, and the blades are widened in step 3. In this way, Cartesian SENSE are adapted to narrow blades.

(3) Both SSB and MJB SENSE are compatible with motion correction. To address more complicated motion, more strategies such as weighting and relaxation, may need to be further developed for MJB SENSE. Note that MJB SENSE can enable higher MB factors, which may promise SMS PROPELLER for true 3D motion correction in the future.

(4) MJB SENSE is efficient, not relying on iterations. Yet it is feasible to combine MJB SENSE with iterations to further improve the reconstructed image quality (shown in another abstract submitted). MJB method may be further applied to GRAPPA-based or non-Cartesian-based methods to enhance SNR and address motion-induced calibration error.