· Fast image acquisition and fast image reconstruction can be used to provide a continuous stream of images (~10frames/s) for procedural guidance in the MRI scanner

· Visualization of devices alongside tissue is essential for procedural guidance.

· Interactive control of image contrast and slice geometry is useful in the interventional MRI environment

Scientists and clinicians interested in guiding invasive procedures using real-time MRI.

The flexibility of MRI contrast makes it appealing for image-guided diagnostic and therapeutic procedures. Standard MRI sequences can provide anatomical, functional and physiological information valuable for pre-procedural planning and intra-procedural imaging. Furthermore, a rapid stream of MR images can be used to guide an invasive procedure in a patient inside of the MRI scanner.

The technical demands of interventional MRI differ from those of diagnostic MRI. To generate images during an interventional procedure, we require fast image acquisition, reconstruction and processing. Moreover, MRI procedural guidance necessitates interactive control of image orientation and image contrast, as well as simultaneous visualization of tissue and interventional devices [1].

Real-time MRI guidance can be used to perform a number of clinical procedures, including tissue biopsy, brain stimulator lead placement, brachytherapy seed delivery, high intensity focused ultrasound ablation and endovascular procedures. This presentation will outline the technology requirements to achieve this real-time MR imaging.

The frame-rate used for MRI-guidance depends on the needs of the procedure. For example, endovascular interventions typically require 5-10 frames/s for safe device navigation, whereas MRI-guided biopsies may use an image every few seconds. MRI must compete with traditional image guidance by X-Ray fluoroscopy or ultrasonography providing upwards of 20 frames/s.

The necessary image weighting and signal-to-noise must be achieved rapidly. For very high frame-rate applications, balanced steady-state free precession (bSSFP) imaging provides the optimal image quality [2], however different image weighting may be preferred for specific applications. Fully sampled Cartesian bSSFP can generate approximately 3 frames/s (eg. TE/TR = 1.27/2.54 ms, matrix = 192x 144). Parallel imaging using GRAPPA or SENSE are commonly used to improve frame rates. Alternatively, EPI and spiral imaging are also appealing to achieve higher frame rates, but require real-time distortion correction [3].

Real-time imaging for procedural guidance is not gated or breath-held. Entire images are acquired as quickly as possible and imaging is run continuously with multiple slices updating in rapid succession (Figure 1).

Magnetization preparations can be toggled on/off to alter image contrast interactively throughout procedure. Commonly, non-selective saturation pulses are added to improve visualization of gadolinium-containing devices [4]. Flow-sensitive preparations can eliminate blood signal while maintaining tissue and gadolinium signal during endovascular procedures [5] (Figure 2). Inversion pulses can be added to provide T1 weighting and improve targeting of pathology [6]. More advanced techniques for cardiovascular procedures include volume selective pulses for MRI angiography [7] and interactive color flow [8].

For interventional MRI, fast image acquisition is only valuable if it is paired with fast image reconstruction. Very little latency between image acquisition and display is acceptable in this setting. Vendor-provided reconstruction software is capable of processing standard Cartesian imaging in real-time, however additional reconstruction tools may be required for non-standard imaging (eg. non-Cartesian imaging, complex parallel imaging schemes).

Coil array compression can be used to reduce the computation burden of reconstruction [9]. Graphics processing unit (GPU) accelerated computing can generate significant improvements in reconstruction time [10], but require additional hardware. External reconstruction computers outfitted with necessary hardware can be used for real-time inline reconstruction using open-source software (eg. Gadgetron [11]). Furthermore, cluster or cloud computing can assist with fast reconstruction [12].

Real-time visualization of interventional devices is also essential for MRI-guided procedures. Device visualization in broadly divided into passive and active visualization. Passive visualization relies on native device properties creating contrast in MRI images. For example, gadolinium filling can be used for catheter visualization [13, 4]. Metallic devices, such as biopsy needles, can be visualized using the signal void in images, emphasized in gradient echo imaging with lengthened TE, or positive contrast methods [14-16].

Active visualization uses embedded receiver electronics to generate a unique device signal that can be separated from the rest of the receiver array. The active device signal can be overlaid in color on MR images, and projection images can be used for accurate device localization when devices move out-of-plane [17]. Point source microcoils require only 3 orthogonal echoes for localization enabling very fast device tracking overlaid on a roadmap images [18].

Typically, MRI-guided procedures require real-time modification of acceleration factor, slice-thickness, image contrast (eg. pulse sequence type, magnetization preparation), as well as graphical slice re-positioning, multislice display, 3D display and color visualization (eg. device tracking, MR-thermometry) [19]. Each vendor offers a real-time imaging environment, or third-party alternatives are available. The typical infrastructure for data acquisition, reconstruction and visualization is shown in Figure 3. A wide variety of pulse sequences and imaging methods are used during interventional MRI, but all applications requires fast image acquisition, fast reconstruction and interactive control of imaging. The imaging technology is available to make MRI guidance achievable in a clinical environment.