SENSE/ASSET

How does SENSE/ASSET work?  

SENSE (SENSitivity Encoding) and ASSET (Array coil Spatial Sensitivity Encoding) are among the most widely used parallel imaging methods. These techniques are primarily performed in image space after reconstruction of data from the individual coils. (This contrasts with GRAPPA/ARC methods which operate primarily on k-space data before image reconstruction).  Each major MR vendor offers some version of the SENSE technique under different trade names: Siemens (mSENSE), GE (ASSET), Philips (SENSE), Hitachi (RAPID - "Rapid Acquisition through Parallel Imaging Design"), and Canon (SPEEDER).

SENSE/ASSET involves 4 steps:

1. Generate coil 
    sensitivity maps

2. Acquire partial k-
    space MR data

3. Reconstruct partial
    FOV images from
    each coil

4. Unfold/Combine
    partial FOV images
    by matrix inversion 

Calculating coil sensitivities are the initial and most important step in the SENSE process. Low-resolution images are acquired separately from each surface coil at full field-of-view. These surface coil images are normalized by dividing them by a low-resolution body coil image. Filtering, thresholding, and point estimation are then applied to the data to generate coil sensitivity maps (shown right). These maps quantify the relative weighting of signals from different points of origin within the reception area of each coil.

The calculation of coils sensitivities may be obtained as a separate ~20 second acquisition before actual imaging begins (GE uses this method for ASSET). Alternatively, automated coil calibration may be integrated into the pulse sequence itself (Siemens' mSENSE). The latter method has the advantage of being less sensitive to motion that occurs between the time of calibration and the beginning of the scan proper.

Once coil sensitivity maps have been calculated, the MR pulse sequence begins. For a PI acceleration factor of 2, alternate lines of k-space are skipped, resulting in a ½-FOV images obtained from each coil with aliasing (wrap-around). A matrix inversion process is used to unfold and combine the aliased images from each coil. How this inversion process works is not as complicated as it may at first appear, so please try to follow the simple 2-pixel example below:

During the prescan calibration step the scanner has calculated point-by-point sensitivities for each surface coil, now stored in memory as a big array of numbers. For an MR signal arising from point A in the patient, the sensitivities of Coils 1 and 2 for detecting that signal will be denoted S1A and S2A respectively. Similarly, the coil sensitivities for any other point B are also known and will be denoted S1B and S2B.

When the data from each coil are reconstructed into images, significant wrap-around (fold-over) artifact is present. This phenomenon, known as aliasing, has occurred because an insufficient number of frequency components have been sampled during the imaging process to uniquely distinguish all spatial locations. Each pixel (P) in the ½-FOV images has a signal that is the sum of contributions from two points (A and B) in the patient. Denoting these pixel values from Coils 1 and 2 by P1 and P2, we can write

P1 = A•S1A + B•S1B
P2 = A•S2A + B•S2B

Since the Pi's and Si's are all known, the true signals (A and B) can be calculated by simple algebraic methods for solving 2 simultaneous equations with 2 unknowns. In the MR scanner a similar process is performed for all data points in the image using a matrix inversion technique, but the idea is the same. Hopefully this example will remove some of the mysteries surrounding the PI reconstruction process for SENSE.

Advanced Discussion (show/hide)»

As you might expect, the reconstruction process is a little more complicated in real life than the above discussion suggests. The typical description for SENSE uses a matrix formulation that includes the following components: I = final image at full FOV; P = partial FOV images with aliasing; S = coarse sensitivity profiles from the individual coils, normalized for uniform signal by the body coil; λ= coarse resolution image from the integrated body coil used to constrain solution during the regularization process; and ψ = noise covariance matrix, representing noise increase due to patient-specific interactions between coil elements.

The inversion process requires solving the following matrix equation using a least squares optimization in the image domain:

I = (S^Hψ^-1S + λ^-1)^-1S^Hψ^-1P

---

References
     Blaimer M, Breuer F, Mueller M, Heidemann RM, Griswold MA, Jakob PM. SMASH, SENSE, PILS, GRAPPA. How to choose the optimal method. Top Magn Reson Imaging 2004;15:223-236 [review]. 
    Deshmane A, Gulani V, Griswold MA, Seiberlich N. Parallel MR imaging. J Magn Reson Imaging 2012;36:55-72. (review)
     Glockner JF, Hu HH, Stanley DW, et al. Parallel MR imaging: a user's guide. Radiographics 2005;25:1279-1297.
     Larkman DJ, Nunes RG. Parallel magnetic resonance imaging. Phys Med Biol 2007;52:R15-R55 [review]
     Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE: Sensitivity encoding for fast MRI. Magn Reson Med 1999; 42:952-962.

---

Related Questions
      Is PI a special type of pulse sequence? Can it be performed with any coil in any direction?
     What is parallel imaging? How does this differ from "regular" imaging?

---

←  Previous Question

Next Question  →

↑ Complete List of Questions ↑