Techniques for fast stereoscopic MRI

Abstract

Stereoscopic MRI can impart 3D perception with only two image acquisitions. This economy over standard multiplanar 3D volume renderings allows faster frame rates, which are needed for real-time imaging applications. Real-time 3D perception may enhance the appreciation of complex anatomical structures, and may improve hand-eye coordination while manipulating a medical device during an image-guided interventional procedure. To this goal, a system is being developed to acquire and display stereoscopic MR images in real-time. A clinically used, fast gradient-recalled echo-train sequence has been modified to produce stereo image pairs. Features have been added for depth cueing, view sharing, and bulk signal suppression. A workstation was attached to a clinical MR scanner for fast data extraction, image reconstruction and stereoscopic image display.

FIG. 1

Modification to (a) FGRE-ET for producing (b) stereo pairs of images. The z (slice-selection) gradient is used during the readout period to rotate about the y (phase-encoding) axis.

FIG. 2

Diagram showing how readout direction is rotated to produce different views of the same slice. Normally, pixel intensities are related to integration across the slice profile, which is normal to the image plane. To produce different views for the left and right eyes, the readout direction is rotated to match the eye separation.

FIG. 3

Some parameters that are specific to a stereoscopic MRI scan. a: Separation angle between the eyes. b: Rotation of viewer and slice position together. c: Rotation of viewer position with respect to the slice position.

FIG. 4

Example stereo pair of images. Some viewers may be able to cross their eyes (without tilting head to either side) to form a third image in the center, which is stereoscopic. The images on the periphery may be blocked with the hands to assist in focusing.

FIG. 5

The effect of the depth-cueing pulse is demonstrated on a constant-signal phantom. The acquisition is rotated such that the slice profile direction is vertical in each image. The viewer position is at the bottom of each image and the left side of each graph. With the RF pulse modified for depth cueing, the image intensity drops with increasing distance from the viewer position. Parameters were b = 1, m = -20, and slice thickness = 500 mm.

FIG. 6

Depth cueing example showing one view of Gd-filled tube. With depth cueing turned on, it is possible to determine which sections of the tube are closer to the viewer. The imaging parameters were: matrix = 256 × 256, flip angle = 20°, ETL = 2, BW = 62.5 KHz, TR = 11 ms, and slice thickness = 500 mm; for depth cueing b = 0.5, m = -10.

FIG. 7

Phantom image with increasing amounts of z-dephasing. Slice thickness = 100 mm; half-cycle distances, Δz, are (in cm): (a) no dephasing, (b) 7.4, (c) 1.29, and (d) 0.488.

FIG. 8

Example of z-dephasing in vivo, Δz = 4 cm, slice thickness = 50 mm, image plane rotated off-axial to show lengths of vessels. Bulk objects are suppressed, easing visualization of smaller objects such as vessels. Each image is scaled to its maximum value.

FIG. 9

Example of depth cueing in vivo. In these off-axial neck images, the more distant tissues appear darker when depth cueing is used. Pay particular attention to tissues traversing the image plane, such as the carotid and vertebral vessels in the dashed region. The imaging parameters were: matrix = 256 × 160, NEX = 1, flip angle = 60°, ETL = 2, BW = 62.5 KHz, FOV = 30 cm, slice thickness = 30 mm, z-dephasing Δz = 7 cm; for depth cueing b = 0.5, m = -1.