MR software tools for real-time decision making and FOV prescription

Introduction

Many cutting-edge MRI neuroimaging paradigms require real-time decision making and precise FOV positioning. We present two software tools, implemented as Siemens image reconstruction modules, to support a variety of such paradigms. The first, called vSend, opens a socket and sends imaging data to another computer where it can be analysed in real-time. The second, called AAhijack, reads a matrix from a socket and overwrites the Siemens AutoAlign matrix, enabling online FOV prescription.

The vSend module can support real-time fMRI (rtfMRI) neurofeedback experiments^1,2 and is compatible with the rtfMRI software murfi2^1,3. It can also be used with a python script⁴ which can read the data sent by vSend and write a NIfTI⁵ file, which allows it to be used in a variety of applications. To date it has been used to: calibrate motion trackers⁶, update shims in real-time⁷, detect fetal head-poses⁸ and determine reacquisition priorities⁹. vSend uses a vendor-agnostic data format¹⁰.

AutoAlign is a Siemens’ tool to automate the alignment of slice positioning in head examinations¹¹. A scout image is acquired from which anatomical landmarks are extracted registered to a template. This creates a 4×4 rigid registration matrix which is saved on the scanner. Subsequent sequences for which AutoAlign is enabled will have their FOV multiplied by this matrix. The AAhijack module opens a socket and queries a remote server for a matrix which it then stores as an AutoAlign matrix. This tool has been used to prescribe FOVs across runs in single-slice BOLD imaging¹².

Methods

Both vSend and AAhijack have been implemented as Siemens ICE functors in IDEA using C++. vSend has been used on the following Siemens software baselines: VB17, VE11C, VE11E, VE12U, XA20A, XA50A. AAhijack has been used on VE11C and VE12U. All results in this abstract were obtained on a 7T Siemens Terra running VE12U.

We validate the vSend module by comparing the NIfTI file derived conventionally from the DICOM images using mri_convert¹³ with the NIfTI file generated via vSend.

We validate AAhijack by using it to store 5 arbitrary AutoAlign matrices and predicting the voxel to RAS (vox2ras) matrices of subsequent AutoAlign-enabled series.

Finally, we demonstrate 2 prospectively updating slice prescriptions systems. The first, called the ‘external method’ (Figure 1), utilizes both vSend and AAhijack. A scout image of a head phantom¹⁴ is acquired and vSend is used to send it to a remote computer. The phantom is then physically repositioned in the scanner and a second scout image is acquired and sent to the remote computer. The computer then registers the two volumes¹⁵ and computes an AutoAlign matrix, which is stored on the scanner using AAhijack. A third AutoAlign-enabled target sequence is acquired and a voxel-wise registration between the first and third image is computed to determine the accuracy of the system. The second, called the ‘internal method’ is like the first, except the scout images are not sent to a remote computer but stored on the scanner and Siemens registration engine (PACE) is used to derive an AutoAlign matrix which is passed to AAhijack.

Figure 1:

Conceptual diagram of the external slice prescription system that uses both the vSend and AAhijack modules. Scout sequences are sent to the external computer using vSend. The external computer registers the scout sequences and computes an AutoAlign Matrix which is sent back to the scanner using AAhijack.

Results

vSend was validated with Gradient Echo and BOLD sequences. NIfTIs generated from DICOM using mri_convert¹³ were compared to the NIfTIs from vSend using mri_diff¹³. The files were identical except for minor differences in the header: TR, TE and flip angle were incorrect in the vSend volume and TE and flip angle were incorrect in the DICOM derived volume. These inaccuracies had no impact on the imaging paradigms that make use of vSend.

We used 5 randomly selected AutoAlign matrices to validate AAhijack. In all cases, the predicted vox2ras matrices were accurate to at least 4 decimal places.

The slice prescription methods were validated visually (Figure 2) and by performing a voxel-wise residual registration¹⁵ between the scout image of the phantom in the initial position and the target image after repositioning the phantom. The residual registration of the external method was 0.0069mm and 0.094 degrees. The residual registration of the internal method was 0.0082mm and 0.14 degrees. The displacement of the phantom was 1.79mm and 3.58 degrees. Figure 3 shows the external method prescribing a single BOLD slice in a human subject as they change position.

Figure 2:

Sagittal, Coronal and Axial slices of the slice prescription systems being used on a head phantom. First row: The scout image in position 1; Second row: The scout image in position 2 after the phantom has been repositioned; Third row: Results of the external slice prescription system; Fourth row: Results of the internal slice prescription system. The displacement is most noticable by observing the nose in the saggital view of the second row, compared to all other rows.

Figure 3:

Saggital view of a single coronal fMRI slice overlaid on a GRE scout. The external slice prescription system is used to update the fMRI FOV as the human subject’s head changes position. Animated GIF, click link to view.

Conclusion and Discussion

The accuracies of the external and internal slice prescription methods are comparable. The internal method doesn’t require external hardware and the external method allows for greater control of the registration process.

The tools presented can be used to support a variety of cutting-edge MR imaging experiments.

Synopsis.

Motivation:

Many cutting-edge MR neuroimaging paradigms require real-time decision making and precise FOV positioning. We present two software tools to support such paradigms.

Goal(s):

Develop two modules.

1. vSend: opens a socket and sends imaging data to another computer in a vendor-agnostic format, enabling real-time analysis.

2. AAhijack: reads a matrix from a socket and overwrites the Siemens AutoAlign matrix, enabling online slice prescription.

Approach:

Modules are implemented as Siemens image reconstruction modules (ICE functors) in C++ and two slice prescription systems utilizing the modules are demonstrated.

Results:

The slice prescription systems have comparable performance and various advantages and disadvantages.

Impact:

The software tools presented have enabled a variety of cutting-edge MR neuroimaging paradigms including real-time fMRI, motion tracker calibration, real-time shimming, fetal head-pose detection and automated FOV prescription, reacquisition planning and single-slice BOLD imaging FOV prescription.