Image-guided surgical navigation systems are designed to help the surgeon in the task of correlating what it is seen in the preoperative medical images and the real anatomy of the patient. The principle behind these systems is that there exists a rigid transformation between the preoperative images and the anatomy of the patient. The process to find this transformation is called “registration” and it consists of selecting at least three corresponding pairs of fiducial points in the preoperative images and in the patient anatomy. In the best case, those fiducials are well known anatomical landmarks, but many times, especially in minimally invasive approaches on complex anatomy, those landmarks are very hard to find. Moreover, there are even more fundamental caveats in the registration process, as those described in the work by Fitzpatrick et al. [3]. In that work the authors mention that point-based registration error can be divided into three different errors:

The main result of Fitzpatrick’s work is the derivation of the statistical distribution for TRE, but the most significant contribution from the application viewpoint is the insight about what is a good registration: a low FRE value is good, but it does not guarantee high accuracy unless other things have been taken into account, like using more than three fiducial points, placing those points far apart and surrounding the target of interest. Another important conclusion is that when “FRE falls below a certain threshold, it gives no further information regarding accuracy”.

A recent study by Stoll et al. [4] in the orthopaedic oncology domain adds a refinement step to the point-based registration. This refinement step generally depends on proprietary surface digitizing devices (surface probing, laser scanning) and computer algorithms that usually perform a small adjustment to the previous fiducial registration step. In their work, Stoll et al. found that even after the surface refinement algorithm provided by a commercial navigation system, the differences between target points and their corresponding points in the navigated images are in a 95 % CI 6.11–16.96 mm.

Multiple forms of evaluating osteotomies have been published, depending both on the intended area of application and on the technology applied to acquire the data used to perform the evaluation. Barrera et al. [5] propose a method to assess the ‘quality’ of bone preparation for knee arthroplasty bone insertion. One of the steps in this assessment process is the estimation of each single planar cut. The method represents the accuracy using five indices for each cut. There is one translational error index (in mm) and three rotational error indices; these last three, combined, result in an overall rotational index (in degrees). The experiments are executed on synthetic knees and the ‘achieved’ surface is digitized after the cut. The article proposes several methods for capturing the surface but it also warns about the different accuracy and precision parameters of those methods.

In a classic article, Cartiaux et al. [6] propose a method based in the ISO 1101:2004 standard for geometrical tolerancing to evaluate differences between a cutting plane and a target plane. Their work shows that it is possible to express the most significant translational and rotational errors using only the location parameter (L) defined in the mentioned ISO standard. This parameter is the maximum euclidean distance from the executed cutting surface to the target plane in a perpendicular trajectory to the last one. For experimental data gathering a test bed with a block simulating bone tissue is used and errors are estimated with a coordinate measuring machine set in the same frame of reference. The method is also used in [7] for evaluating different bone cutting technologies. In this work the error (L) was 0.92 ± 0.37 mm with a robot-assisted process compared with 1.26 ± 0.88 mm with the freehand process (p < 0.0001) and 1.87 ± 2.09 mm with the navigated freehand process (p < 0.0001).

So et al. [8] introduce a new registration procedure using fluoro-CT matching and evaluate its accuracy using the postoperative gross measurement of surgical margin. The accuracy in this case is related to the planned margin, and it is not an absolute value.

Dobbe et al. [9] propose a method to measure and estimate the normal of an executed plane. This normal is used to compute the dihedral angle with the target plane, that is decomposed in sagittal and coronal plane angles. Then a distance error between the target and executed plane is computed taking the Euler distance between the centroids of the cross sections defined by target and executed planes. This method is validated using a cadaveric limb, with pre and postoperative computed tomography (CT) scans positioned in a common frame of reference using a registration algorithm. The methodological accuracy and precision is also evaluated, showing that the method introduces an error that it is well below 0.5 mm in mean.

Stiehl et al. suggest [10] using tools borrowed from the field of statistical process control in the domain of accuracy and precision evaluation in computer-assisted orthopaedic surgery. Milano et al. [11] follow that path introducing a definition of accuracy and precision that could be used with the industry proved process performance index as a clinical score and using CT scans to digitize the surgical specimen resected from the patient. The introduction of an index helps to avoid the problem of measuring the accuracy against a fixed frame of reference. This work also evaluates the methodological error. This error is below 1 mm in all the test cases. A first application of this methodology in the evaluation of surgical accuracy in 61 osteotomies performed on 28 patients is found in the article by Ritacco et al. [12]; this work shows that the accuracy parameter is 2.52 ± 2.32 mm.

In a recent article, Sternheim et al. [13] use a custom navigation system with synthetic and cadaveric pelvic bones to generate a large number of cuts. The cuts are evaluated by CT scanning the bones after the osteotomies and measuring the entry and exit cut distances and deriving the pitch and roll angle differences. In navigated cuts, using synthetic bones, the entry error is 1.6 ± 1.1 mm and the exit error is 2.3 ± 1.1 mm, while in non-navigated cuts the entry error is 2.8 ± 4.9 mm and the exit error is 3.5 ± 4.6 mm.

Patient-specific instrumentation (PSI) technology is an alternative to intraoperative navigation. The accuracy of PSI technology adapted for bone tumor surgery has been studied by Cartiaux et al. in [14]. In that article experiments are conducted using synthetic right hemipelvic bone models that are fixed to a test bed, setting a global reference frame for measurement. The test bed is digitized using a CT-scanner and a simulated tumor is introduced in the bone. A mixed seniority group of 24 orthopaedic surgeons performs the bone cuts with a pneumatic oscillating saw. The experiments show that the location accuracy of the cut planes varies significantly in terms of mean and 95 % confidence interval (CI) among the four target planes. The average location accuracy in the anterior and posterior ilium is 1.0 mm (CI 0.8–1.3 mm) and 1.2 mm (CI 0.9–1.6 mm) respectively and it is significantly different from the average in the pubis and ischium, 2.0 mm (CI 1.5–2.7 mm) and 3.7 mm (CI 2.8–4.9 mm) respectively. The surgical margins achieved in the pubis, with an average of 11.8 mm (CI 11.3–12.3 mm), were significantly higher than those achieved in the ischium and anterior and posterior ilium, with an average of 9.2 mm (CI 8.6–9.7 mm), 10.0 mm (CI 9.5–10.6 mm) and 9.7 mm (CI 9.2–10.3 mm) respectively.