Principles and concepts of surgical navigation





Computer-assisted surgery consists in using the calculation power, data storage and graphic display capabilities of microcomputers to make a real-time match between the spatial coordinates of sensors affixed onto an instrument used by the surgeon and the corresponding points on patient imaging. The result is displayed on-screen, visible to the surgeon, who thus has the illusion of navigating among the patient’s anatomic structures. The position of the instrument as surgery progresses is exactly known, even in areas that are narrow, hemorrhaged, or remodeled by disease or by fibrosis induced by previous surgery.


The technology thus presupposes a spatial localization system, a computer and adequate CT or MRI imaging ( Figure 2.1 ).




Figure 2.1


General principle of surgical navigation.

(Source: Bertrand Lombard.)


Spatial localization


Spatial localization is the essential element in this technology, converting the position of a sensor on the instrument into relative Cartesian coordinates.


Electromechanical systems


Electromechanical systems were the first to be used, being easy to construct. A jib supports an arm comprising several segments articulated to provide 5 or 6 degrees of freedom ( Figure 2.2 ). The articulations usually include a precision potentiometer encoding the angle between each segment. Accuracy was to within a few micrometers, but they required the patient to be totally immobile. Other reasons for their being abandoned were bulk and poor ergonomy.




Figure 2.2


An electromechanical localizer.

(Source: Bertrand Lombard.)


Ultrasound-beam systems


Ultrasound-beam systems comprise a piezoelectric ultrasonic transmitter fixed on the instrument, generating a signal picked up by a set of at least three microphones remote from the operative field. If sound propagation time in air is known, the relative position of the instrument can be calculated by quadratic triangulation. Unfortunately, sound propagation time is highly dependent on temperature and local hygrometry, leading to cumulative error of as much as several centimeters, even after meticulous calibration. Moreover, interposition of the surgeon’s hands and instruments and any sudden displacement of air induce errors that are hard to detect. These systems have therefore remained at the prototype stage.


Opto-electronic systems


Opto-electronic systems were developed a little later. They are based on a set of two or three high-resolution charge-coupled cameras remote from the operative field. They analyze the spatial projection disparity of an identifiable object such as a set of near-infrared LEDs fixed on the instrument (active system), or microprism-covered spheres reflecting infrared emission from a projector in the camera bay (passive system), so that no cable is needed to feed the LEDs ( Figure 2.3 ). Accuracy is satisfactory, at around half a millimeter for electroluminescent diode systems, and slightly poorer for passive systems.




Figure 2.3


Working principle of an optical tracker.

Two cameras ( blue ) with concentric infrared (IR) LED projectors ( red ) observe the tracking area illuminated by the IR light. The object ( yellow ) reflecting IR light projects its position on the image plane of each camera, creating two sets of 2D coordinates (x 1 , y 1 ); (x 2 , y 2 ). Due to the disparity distance d , the 3D coordinates of the object may be computed, knowing the focal length of each camera (i.e. their image-plane location).

(Source: Bertrand Lombard.)


Some systems use a geometric or colored marker on the tracked instrument. The signal from one or several cameras is processed by a pattern-feature recognition algorithm to search for the marker structure and extract it within the image flow. After extracting the center of each marker, an algorithm determines the spatial position of the target and thus of the instrument.


All optical systems have the drawback of being interrupted whenever an object gets in the line of sight between the camera and what is supposed to be tracked. This makes ergonomics difficult to manage, as the surgeon is not free in positioning aids, instruments and hands. This is particularly critical in operating in deep narrow areas, as in transnasal pituitary adenoma resection. The camera bay is bulky and, above all, fragile, as it contains a system of cylindrical lenses liable to lose focus under shock, impairing accuracy. If these systems are finely suited to the more-or-less static procedures typically encountered in neurosurgery (for which they were originally designed), they are much less suited to dynamic surgery such as ENT craniofacial procedures.


Electromagnetic systems


Electromagnetic systems are a more recent development. They are based on generating an alternating magnetic field on three perpendicular axes. A sensor, comprising coils that are also perpendicular to one another, detects its position and orientation relative to the transmitter according to the voltages induced in each coil ( Figure 2.4 ). As magnetic field strength decreases with the cube of the distance from the source, sensitivity and resolution are excellent, to within a 10 th of a millimeter. The technology is very robust, and can be miniaturized to the point where the surgeon can just forget about its presence. Another advantage is that they function without restriction, whatever the position of the surgeon, instruments and aids, as they do not depend upon a line of sight.




Figure 2.4


Principle of an electromagnetic localizer.

Three AC wave generators supply current to 3 perpendicular coils which in turn generate a magnetic field looped in the 3D space, according to Maxwell’s 1 st law. These fields induce a current in the sensor, also comprising 3 perpendicular coils. Their weak signals are processed and an algorithm estimates the position and orientation of the sensor relative to the transmitter frame. Using 3 frequencies (of the order of a kiloHertz) creates 3 separate channels. Other channel multiplexing designs or coil arrangements are possible.

(Source: Bertrand Lombard.)


The transmitter, measuring 1 or 2 cm 3 , can be fitted directly on the patient, without the kind of reference sensor needed by infrared systems, thus halving the risk of measurement error. The early versions had the major inconvenience of being sensitive to metallic objects, inevitably present in the operative field; this has now been minimized by sophisticated signal generation and processing technology eliminating most metal-induced distortion. This electronic technique can be further combined to algorithms able to predict onset of distortion to a high degree of probability and to correct any such error using a predefined mathematical model such as the finite elements method [ ]. Associating an extra control sensor further increases reliability, warning the user of onset of uncorrected error. This technology now ensures reliability at least as good as that of infrared systems, without their drawbacks.


Other tracking technologies


Gyroscopic inertial sensors combined to accelerometers or spectrum-modulating fiberoptic sensors (in which transmission varies with fiber curvature radius) have been described, but are not widespread due to specific drawbacks. On the other hand, direct intraoperative cone-beam X-ray or intraoperative MRI imaging are feasible, enabling direct image analysis without necessarily needing a spatial localizer. Although costly and not easy to implement, these techniques provide real-time up-dated imaging, enabling tissue margin assessment during resection.


Image dataset registration


Principle


The position and orientation of the patient’s head on imaging depends on how he/she was installed on the radiology table with respect to the radiogenic tube. There is obviously no reason why, once positioned lying on the operating table, the patient’s position should remain identical with respect to the localizer. Registration therefore seeks to determine transfer equations to convert the coordinates of any point on the localizer to match exactly the corresponding point in the image coordinate system. These transfer equations from one coordinate system to another generally take the form of a rotation matrix around three axes (x, y, z), combined with a translation vector along each axis.


Paired points registration


The simplest registration technique is to provide the machine with the coordinates of a certain number of points acquired on the patient’s face and skull and select the corresponding points on the images ( Figure 2.5 ). Natural anatomic landmarks, such as the nasion, frontozygomatic suture, tragus or nasolabial junction, can be used, or else radio-opaque markers fixed on the skin or screwed onto external skull.




Figure 2.5


Paired points registration.

(Source: Bertrand Lombard.)


The latter technique is more demanding for both patient and physician, requiring a specific radiologic examination, to be followed by surgery as quickly as possible to avoid the discomfort of the screwed markers or displacement of glued markers. But it is more precise than the former, not depending on the surgeon’s ability to locate a 3D point on imaging and to target exactly the same point on the patient.


Surface registration


The principle of surface registration is to create a 3D surface model, with predefined resolution, from a set of images, then to acquire hundreds of random points on the patient’s face via a probe governed by the localizer, and perform iterative rotation and translation until optimal superimposition on the surface model is obtained by monotonous convergence. Algorithms prevent the convergence being trapped in a local minimum rather than the global minimum of the function. Adjustment requires only a few seconds’ calculation, performed automatically by the computer ( Figure 2.6 ).




Figure 2.6


Surface registration.

A. Patient’s randomly recorded points. B. Automatic adjustment on patient’s image set.

(Source: Bertrand Lombard.)


This technology has several important practical advantages:




  • no dedicated radiologic examination is needed ahead of surgery;



  • the procedure is simple, very quick and relatively operator-independent;



  • point-matching is theoretically several thousand times more precise than when manual;



  • and, above all, the number of points used for registration can be far greater than the 5 to 10 points used in conventional methods. Since the mathematical optimization of registration is a function of the square root of the number of sampled points, accuracy may be significantly improved.



Use of a laser further simplifies the procedure, but is restricted to perfectly hairless skin areas [ ].


Photo- and video-based registration


Photogrammetric methods have been proposed in order to make the image-to-patient registration process even simpler. Photogrammetry, which is widely used in geodesic science, for instance to create maps from high-resolution satellite pictures, relies generally on 2D-Fourier image decomposition to extract spectral content and hence the most significant signals. If the camera is 3D-tracked, it is possible to a certain extent to compute the relative translation and rotation of broadly similar pictures. For surgical purposes, it is possible to use this image processing technology to find the pose of a surgical view relative to the 3D reconstruction performed from the patient’s CT dataset [ ]. Despite their appealing potential, these techniques are very difficult to apply to surgical registration, due to lack of robustness. With the same purpose, some authors have proposed using endoscope image streams to dynamically register the patient relative to the image dataset. While some interesting results have been demonstrated in lab conditions on immobile dry plastic skulls [ ], such methods are hard to apply under real surgical conditions where blood, tumor and, moreover, the progressive modification of the surgical field by the surgical procedure itself, seriously vitiate success.


Mathematical limits of registration


Whatever the registration method, it has to take account of the fact that matching the patient images (model) and patient data is inevitably subject to noise, precluding any formal equation. The noise comes from the patient’s movements during examination, imperfections in the imaging device, morphologic modifications involved in patient positioning on the table [ ], localizer imprecision, and of course possibly imperfect matching of coordinates. Numeric optimization techniques are therefore used, to minimize noise in the data. The simplest is affine registration, which, by not requiring a strictly orthogonal Cartesian space, may allow poorly sampled data to be matched.


Estimating registration accuracy


Registration accuracy is obviously a very useful item of information, but cannot be assessed other than statistically, as the very principle of registration involves a relation between two coordinate systems that is unknown a-priori and for which no ground truth is available. Mathematically, it is generally estimated from the standard deviation of the geometric disparity of each registration point, once transformed by the registration ( eq. 1 ):


<SPAN role=presentation tabIndex=0 id=MathJax-Element-1-Frame class=MathJax style="POSITION: relative" data-mathml='Er=∑​1i=nPim−T,R,Pil−μn−1′>Er=𝑖=𝑛1((𝑃𝑚𝑖(𝑇,𝑅,)𝑃𝑙𝑖))𝜇𝑛1Er=∑​1i=nPim−T,R,Pil−μn−1
Er=∑​1i=nPim−T,R,Pil−μn−1

Only gold members can continue reading. Log In or Register to continue

Stay updated, free articles. Join our Telegram channel

Oct 27, 2024 | Posted by in OPHTHALMOLOGY | Comments Off on Principles and concepts of surgical navigation

Full access? Get Clinical Tree

Get Clinical Tree app for offline access