Intraoperative ultrasound (US) in image-guided surgery (IGS)

• Overview

Intra-operative imaging has been dominated by fluoroscopy for many years, followed more recently by mobile computed tomography (CT) platforms. However, both modalities make radiation exposure a concern for patients and surgeons alike. In the last several years, magnetic resonance imaging (MRI) has been finding its way into the operating room (OR), making high-contrast soft-tissue details available during surgery, but the size of these MRI machines makes them bulky for use in an already crowded OR. Moreover, MRI still remains far more expensive, severely restricting its use. Although ultrasound has established itself as an important imaging method for applications like obstetrics and breast imaging, its use as a guidance tool has been fairly limited in the past. Compared with other modalities, ultrasound yields images that are lower in quality for surgical guidance. Also ultrasound lacks of registration coordinates, owing to the freehand nature of ultrasound image acquisition. However, digital beam formers and harmonic imaging have greatly improved ultrasound image quality. Extensive research is being conducted to tackle the speckle that plagues ultrasound images. Motorized transducers for volume data acquisition and advanced rendering algorithms have made 3-D ultrasound a reality. New motion tracking system has been made available in the OR. These advances, combined with the inherent safety and real- time nature of ultrasound, indicate that ultrasound could be a viable modality for surgical navigation. Up till now, ultrasound guidance has been successfully used for areas like breast biopsies and other tissue aspiration for almost a decade. More recently, ultrasound- guided injections have proved effective for treating musculoskeletal conditions like injured tendons, ligaments, and joints. Intravascular ultrasound is also becoming valuable in stent placements. By visualizing stents in real time, surgeons can position them precisely, thereby reducing occurrence of restenosis. With the advent of portable handheld equipment that lends itself to the space limitations within the OR, ultrasound is even more attractive for intra-operative use.

• Ultrasound based image-guided surgery system

Generally speaking, there are four basic components for the IGS system: image acquisition, 3D tracking, image-to-physical-space registration, and display of image and location. In addition, some application also requires image segmentation, where pathological tissues or anatomical structures are identified on the images. However, for ultrasound image segmentation, the results are far less accurate than MRI and CT scans. Among all the components, the tracking system is an important link between the image and physical space. The old approach employed a probe that was physically linked by a multi-joint arm (called Faro arm) attached to the apparatus restraining the body, and the position was taken by computer from the angles of each joint. However, this approach had restricted the freedom of movements for the tracking probe. To gain more flexibility, the mostly used tracking system employs optical approaches, where an optical tracker emits infrared light on the targets, and then receives back the infrared rays reflected by the reflective spheres mounted on the tracking targets. Information about position and orientation of the targets are saved in a coordinate system relative to the world space (the patient space). The computer will then display the tracked targets with respect to the pre-operative images, in order to guide the surgical tools according to the patient anatomy and pre-surgical plans. In practice, these devices can achieve a tracking accuracy of at least 0.5 mm (Wiles et al 2004). Further more, to track the position and direction of a tool deeper inside a body, a real-time imaging system is often required, and in many cases, ultrasound serves a good candidate.

•  Image registration for ultrasound images

In order to properly guide the surgical tools from real-time images, the position and orientation of tracked targets needs to be brought from the world coordinates to the coordinates of pre-operative images, or sometimes vice versa. This process is a classical problem of coordinate transformation. All components shown in figure 2 are linked relative to each other in space by transformation matrices (3D translation and 3D rotation), which are obtained in a calibration process through matching the pairs of reference points in patient space and the ones in pre-operative images, followed by an application of the least-squared minimization techniques. The tracking system and coordination transform is helpful to bring physical structures together in one coordinate system assuming that all elements are rigid bodies, but to learn the up-to-date morphological information of soft tissues and therefore more precise location of surgical tools, there is a need of image registration, which aligns multiple images (e.g. ultrasound image and MRI) through matching landmarks or common features, linearly or non-linearly. To borrow the real-time imaging ability of conventional ultrasound and high spatial resolution of pre-operative image data, medical image registration has become highly desirable in many IGS’s. By applying and extending regular 2D image registration, commonly seen in the study of computer vision, medical images of different modalities and different dimensions can be semi-automatically or automatically aligned together, spatially matching the anatomical structures. In the current literature, ultrasound image registration has been investigated for the purposes of multiple types of surgeries, including biopsy, neurosurgery, spinal surgery, and so on.Common registration methods for medical images are grouped into two categories, feature based and intensity based. Feature-based registration relies on the presence and recognition of landmarks or fiducial markers to judge the goodness of alignment; Intensity –based registration measures directly the similarity between pixels or voxel intensities of the entire images. Due to the drawbacks of the first kind, such as inconvenience of planting landmarks, difficulty of determining satisfactory fiducial markers and dependency on segmentation procedures that can introduce errors, the latter category has been put under more in-depth study. For this category, there are many metrics to assess the quality of image alignment through intensity map similarity measurements. As reported by L. Zollei et al (2001), by experimenting with six target functions (normalized correlation, entropy of the difference image, pattern intensity, mutual information, gradient correlation and gradient difference),the best metrics of similarity measurements for 2D and 3D multi-modality registration are pattern intensity and gradient differences.