### 2857

FPGA based real-time sensitivity maps estimation using pre-scan method.

### Synopsis

Accurate estimation of the receiver coil sensitivities is critical for an error-free image reconstruction from under-sampled data in SENSE. This work proposes an FPGA (Field Programmable Gate Array) based application specific hardware, for real-time sensitivity maps estimation using pre-scan method. In the proposed work, SENSE reconstructions are performed using the sensitivity maps (computed from the proposed design) and the under-sampled data. The results show that the proposed architecture computes receiver coil sensitivity maps in only 1.466 ms for 8 receiver coils. Also, SENSE reconstructed images show a good mean SNR (30+dB) and low artefact-power (<6×10-4).

### Purpose

Estimating the receiver coil sensitivity maps in real-time without reducing their accuracy by implementing the pre-scan method of sensitivity estimation on FPGA.

### Methods

Parallel MRI (pMRI)1 provides a framework for reducing the MRI scan time by acquiring a reduced amount of k-space data with an array of multiple independent receiver coils. This under-sampling leads to aliased images. Artefact free images can be reconstructed from the under-sampled data using SENSitivity Encoding (SENSE)2, a pMRI method for image reconstruction in image domain. SENSE works by multiplying the inverse of the sensitivity encoding matrix with the aliased images. For an error-free reconstruction, sensitivity maps of the receiver coils must be accurately known. In the pre-scan method of sensitivity maps estimation, a quick separate scan is performed right before the actual MRI scan to acquire low-resolution images (of the object to be imaged) using all the receiver coils. These low-resolution images are used for computing a sum of squared image, and its square-root is calculated. Then the low-resolution images are divided by the square root of the sum of squared image to get sensitivity maps2.

Parallelism in the steps of the pre-scan method can be exploited to significantly reduce the map estimation time. Recently, an FPGA based implementation of SENSE3,4 has been proposed but it does not include the estimation of receiver coil sensitivity maps on the FPGA architecture. This paper proposes an FPGA based solution of pre-scan sensitivity maps estimation method. In the proposed architecture, FPGA block RAMs are used to store the low-resolution data acquired from the MRI scanner. Then, the proposed architecture performs the pre-scan method steps to produce the sensitivity maps which are then stored back in the FPGA block RAMs. Finally, the output is transferred from FPGA to MATLAB using Universal-asynchronous-receiver-transmitter(UART) serial communication for validation of the results.

### Results

Eight channel receiver coil datasets (phantom and human head) with an acceleration-factor of 2, are acquired using GE MR450, 1.5T MRI scanner using Gradient Spin Echo sequence with the following scan parameters: Slice thickness $3mm$, Matrix size $256\times256$, Flip Angle $90°$, TR $520ms$, TE $15ms$ and FOV $55mm$. Figure 1 and Figure 2 show a comparison of the sensitivity maps estimated by pre-scan method using MATLAB and the proposed architecture for the phantom and human head images, respectively. Figure 3 shows the images reconstructed using the sensitivity maps estimated by the proposed system and MATLAB. Table 1 compares the SNR and artefact power values of the reconstructed images. In Table 2, a computation time analysis of the proposed architecture and MATLAB for pre-scan method is presented, which shows that the proposed system takes only $1.466ms$ to compute the sensitivity maps for human head and phantom data sets (8 receiver coils) whereas the MATLAB takes $419.2ms$.

### Discussion

The difference images in Figure 1 and Figure 2 show that the maps estimated by the proposed architecture are very similar to the maps estimated by MATLAB. Furthermore, the images reconstructed using the sensitivity maps estimated by the proposed method are also similar to the images reconstructed using the sensitivity maps calculated by MATLAB. Table 1 shows that the images reconstructed using the sensitivity maps estimated by the proposed architecture provide low values of artefact power $(<6\times10^{-4})$ and good mean SNR $(30+dB)$.

The proposed architecture computes the sensitivity maps almost 400 times faster than the sensitivity maps estimated using MATLAB (running on Intel(R) Core™ i3-4010U CPU @1.70GHz). The proposed method takes only $1.466ms$ for the estimation of sensitivity maps for matrix size $256\times256\times8$ as compared to $419.2ms$ on MATLAB.

The proposed design can be used to estimate the sensitivity maps without the need to transfer raw complex data (all sensitivity maps and aliased images) to the memory banks of the MRI workstation, thereby, reducing the coaxial noise. Real-time sensitivity maps estimation can also help in applications where the sensitivity maps need to be acquired again and again like in cardiac MRI, and it would also help to decrease the overall image reconstruction time in SENSE.

### Conclusion

An FPGA based system for sensitivity maps estimation using pre-scan method is proposed. This system speeds up the process of sensitivity maps estimation up to 400 times (than multicore-CPU) while maintaining quality of the reconstructed images i.e. artefact power $<6\times10^{-4}$ and mean SNR $30+dB$. Accurate and real-time estimation of sensitivity maps right on the receiver coil data acquisition system, makes the system memory and transmission cost efficient, which can be very useful for modern MRI systems.

### Acknowledgements

No acknowledgement found.

### References

1. Larkman DJ, Nunes RG. Parallel magnetic resonance imaging. Physics in Medicine and Biology. 2007;52(7):R15-55.

2. Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE: sensitivity encoding for fast MRI. Magn Reson Med. 1999;42(5):952-62.

3. Siddiqui MF, Reza AW, Omer H, et al. Parameterized architecture design of SENSE for real-time reconstruction. Magnetic Resonance Materials in Physics, Biology and Medicine. 2015; 28, Supplement 1:277-418.

4. Siddiqui MF, Reza AW, Shafique A, Omer H, & Kanesan J. FPGA implementation of real-time SENSE reconstruction using pre-scan and Emaps sensitivities. Magnetic Resonance Imaging. 2017; 44:82-91.

### Figures

Figure 1: Comparison of the pre-scan sensitivity maps estimated for phantom images. (i) Sensitivity maps estimated by the proposed architecture, (ii) Sensitivity maps estimated by MATLAB, (iii) Difference between the sensitivity maps estimated by the proposed architecture and MATLAB

Figure 2: Comparison of the pre-scan sensitivity maps estimated for human head images. (i) Sensitivity maps estimated by the proposed architecture, (ii) Sensitivity maps estimated by MATLAB, (iii) Difference between the sensitivity maps estimated by the proposed architecture and MATLAB

Figure 3: Reconstructed images (phantom and human head) using sensitivity maps estimated by (i) MATLAB and (ii) Proposed FPGA architecture and under sampled data at acceleration-factor 2

Table 1: Comparison of the reconstructed images (phantom and human head) in terms of artefact power and mean SNR

Table 2: Computational time analysis for sensitivity maps estimation, for 8 receiver coils, using the proposed architecture and MATLAB (running on Intel(R) Core™ i3-4010U CPU @1.70GHz)

Proc. Intl. Soc. Mag. Reson. Med. 26 (2018)
2857