The reliability of frame-based stereotactic biopsy
Overall, our data support the notion that stereotactic biopsy is an effective means of establishing tissue diagnosis for intracranial lesions. Our diagnostic yield of 94% compared favorably with recent and historical large series reporting ranges of 90–98% (Table
5). Even for the 6% of our cases with nondiagnostic results, tissue pathology was still suggestive enough to effectively guide treatment in all but one instance. In our series, lesion size and/or location did not correlate with diagnostic yield or accuracy.
Table 5
Summary of large stereotactic series
| University of Friedburg | 1980 | Reichart Mundinger CT indexed to intra-operative ventriculogram | 302 | >90 | NR | 2.3 | 3 |
| Karolinska | 1981 | Leksell | 345 | 91 if ct used | ? | <1 | 2.3 |
| Madrid | 1982 | Leksell | 100 | 97 | NR | 7 | 0 |
| USC | 1983 | BRW | 83 | 94 | NR | 4 | 0 |
| Pittsburgh | 1984 | Leksell | 102 | 96.5 | NR | 3 | 0 |
| Friedburg | 1985 | Mundinger | 815 | | | 3 | 0.6 |
| USC | 1987 | BRW | 500 | 95.6 | ? | 1 | 0.2 |
| Toronto | 1986–1994 | BRW | 300 | NR | NR | 6.3 | 1.7 |
| JGU | 1994 | BRW/CRW | 200 | 92 | 91 | 3 | 1 |
| Centre Hospitalier Universitaire la Timone, Marseille, France | 1996 | 15centers: Talairach 8 Leksell 2 both 2 other 3 | 370 | 94 | 97.7 | 0.8 sig. 7 transient | 1.3 |
| Gainesville | 2001 | | 200 | 98.5 | | 2 | 0 |
| Pittsburgh | 2001 | Leksell | 500 | NR | NR | 1.28% radiographic hemorrhage | 0.2 |
| UCSF | 2006 | CRW or BRW | 213a
| 90 | NR | 2 | 0 |
| Johns-Hopkins | 2006 | Leksell or CRW | 160a
| 91 | NR | 13 | 1 |
Linskey (present) | U. Arkansas | 2007 | CRW or Leksell | 106 | 96 | 91 | 4 | 0 |
Diagnostic accuracy is an equally important consideration in evaluating the utility of stereotactic biopsy, though it is reported far less frequently in the literature than diagnostic yield. At least one study has suggested that biopsy specimens cannot provide a sufficient accuracy of diagnosis to reliably guide treatment of brain neoplasms, citing discrepancies as high as 38–49% when biopsies specimens were compared with final pathological diagnosis obtained at open surgical resection [
14]. Those figures, however, were based upon biopsies performed by multiple outside facilities and few details regarding the specific techniques used are provided. Large single center series with higher individual procedure volumes are likely more representative of the true potential of stereotactic biopsy, and a thorough and systematic biopsy technique is necessary for optimal results. Our accuracy rate of 90.9% is in keeping with that of Grunert et al. [
15], who reported a diagnostic accuracy of 91% in 47 patients who underwent a subsequent open resection. Woodworth et al. [
16] reported an accuracy rate of 76%, though results correctly guided treatment in 91% of their series of 21 patients who underwent open biopsy. Our single inaccuracy was a WHO II oligodendroglioma which was subsequently found to be WHO III at the time of resection. While we did not detect this tendency, a few reports have suggested that mixed gliomas with a significant oligodendroglial component may be more commonly mis-graded on stereotactic biopsy than other glioma histologies [
14,
16].
The specific method of tissue biopsy likely plays a key role in determining diagnostic accuracy and yield. Past series appear to be relatively evenly divided between the use of biopsy forceps [
1,
4] or a side-cutting biopsy needle [
17,
18] while many report using both [
15,
19,
20], and still other surgeons still use needle aspiration techniques or the Backlund spiral devise (Elekta, Inc, Norcross, GA). More recent series tend to favor a side-cutting needle exclusively [
10,
11,
18], which has the advantage of preserving a core of intact cross-sectional tissue architecture which facilitates histological interpretation. The present series used a relatively aggressive biopsy technique which we feel minimizes sampling error and increases the likelihood of an accurate diagnosis. When practical, a target point beyond the edge of the lesion was selected. Multiple sections were then taken with the side-cutting needle at serial depths along the track. In this manner a “geologic core” could be obtained with a single needle trajectory, providing samples of normal brain, lesion edge, and central contents. The utility of this approach is reflected in the accurate grading of all but one of the gliomas in our series, and in the low number (three) of necrosis-only results in our GBM biopsies.
Role of stereotactic biopsy in clinical decision making
One particularly important role for stereotactic biopsy is confirming tissue diagnosis for patients with multiple brain lesions, specifically in the setting of a negative systemic metastatic survey. Of our 25 multifocal lesions, a significant majority (68%) were primary CNS neoplasms, reinforcing the importance of tissue diagnosis when no obvious metastatic source can be found. Not all patients with multiple CNS brain lesions can be assumed to have metastatic disease. Even higher instances of multifocal primary CNS disease have been reported elsewhere. Yamada et al. [
21] found zero metastases out of 25 multifocal brain lesions that were referred for stereotactic biopsy. All three patients with multifocal lesions described in Lunsford et al. [
6] were found to have gliomas. In all, Lunsford also found that nearly 10% of pre-procedure diagnoses classified as “secure” were overturned after biopsy. Another nine out of 44 patients with a “strongly suspected” pre-op diagnosis were found to have a pathology that was not considered in the pre-biopsy differential. A retrospective review by Arbit and Galicich [
22] similarly found that results of stereotactic biopsy dictated different treatment than radiographic diagnosis in 19% of cases.
Our data also indicate that stereotactic biopsy is an important tool for establishing the diagnosis of corpus callosum lesions. Conventional teaching has often been that patients with lesions crossing the corpus callosum do not have resectable lesions and can be assumed to have GBM, and thus can be empirically treated. While most of the 18 patients in our series with a callosal lesion did turn out to have GBM, 27.8% had lesions which mandated different management, including oligodendroglioma, PCNSL, and tumescent MS. This finding becomes particularly important given that some authors have questioned the utility of biopsy in the management of gliomas [
14]. In that study, only three out of 81 lesions were located in the corpus callosum. The present data indicate that a diagnosis of glioma based solely upon characteristic imaging is premature without a tissue diagnosis.
Safety considerations in stereotactic biopsy
It is of interest that the morbidity rates reported in recent stereotactic series differ little from those performed nearly three decades ago (Table
5). Mortality rates, in contrast, have tended to decline slightly over the same period, possibly through technological and infrastructure improvements which allow for faster recognition and correction of post-procedural emergencies. Our morbidity rate of 4% (temporary or permanent neurologic deficit) corroborates the results of these other large series and also demonstrates that an aggressive sampling technique can be employed without compromising patient safety. This is consistent with the study of Mcgirt et al. [
23], which found that increasing the number of biopsy samples did not independently impact morbidity if the samples were collected along a single needle trajectory.
Brainstem and pineal locations accounted for three of four complications, and this is consistent with the findings of previous authors who correlated morbidity with pineal [
18] and deep-seated lesions [
23]. In general, relatively small numbers of pineal locations in this and other series make generalization to an accurate risk profile for these lesions difficult. A contrary view was offered by Regis et al. [
19], who reported the results of a multicenter series of 370 pineal region stereotactic biopsies and found that complication rates were no higher than in other locations if only permanent deficits were considered. That study did note an increased likelihood of complication associated with “hard” tumors (pineocytomas, teratomas, and astrocytomas) and recommends proceeding with a microsurgical approach in the event that tissue is not easily obtained with the first pass of the biopsy needle.
In seeking possible predictors of post-procedure complication, we have specifically identified the finding of blood within the biopsy needle that persists beyond two needle irrigations. Shastri-Hurst et al. [
24] have previously noted the finding of blood intra-operatively in 7/203 cases as having a positive predictive value of 57% for post-operative deterioration but a sensitivity of only 30%. In our series, 19% of patients met our particular criteria for persistent intra-operative bleeding. We found the absence of this finding in the remaining 81% to have a very high (98.6%) negative predicative value for a significant (>5 mm) hemorrhage being identified on the post-operative CT. Our analysis of the Shastri-Hurst et al. data reveals a similar negative predicative value (95.4%) for post-procedural deterioration. With such high negative predicative values, routine post-biopsy neuroimaging in the absence of bleeding through the needle persisting beyond two irrigations or development of a new neurological deficit can probably be safely eliminated from stereotactic needle biopsy patient care protocols.
While some authors have suggested that patients with normal postoperative scans do not require further assessment [
25], Field et al. [
18] found a small but non-zero incidence (0.4%) of delayed neurologic deterioration after uncomplicated brain biopsy with negative post-operative imaging. This leaves the question of an appropriate level of nursing care for these patients still open to debate. It is our opinion that a step-down level facility or well trained neurosurgical floor capable of providing neurologic assessments every 2 h is sufficient for the initial 12 post-operative hours in uncomplicated cases. In other publications, poor glycemic control in diabetics [
23] and platelet counts <150,000 [
18] have also been identified as statistically significant independent risk factors for poor outcome after stereotactic biopsy and should be considered in post-biopsy management decisions. .
Frameless stereotaxy: a hypothetical cohort
Given the increasing popularity of frameless neuro-navigation systems in stereotactic biopsy, we sought to establish criteria through which we could determine the suitability of frameless stereotaxis (FL) for patients in our series. We began with the assumption that frame-based stereotaxis (FB) represents the gold standard for targeting accuracy and examined the literature for assessments of FB and FL systems.
In reviewing the spatial accuracy of various FB and FL systems, it is important to understand the practical significance of the different measures of accuracy commonly cited. Intraoperative computer workstations provide an estimate of root mean square error (RMS) following co-registration of skin fiducials or the stereotactic headframe. However this value should not be considered indicative of true accuracy. Rather, RMS represents the degree of internal consistency between data points––in this case the computed coordinates within the virtual space of the computer workstation. RMS gives no information regarding the correspondence of those coordinates to the actual location of objects in physical space. This concept was elegantly demonstrated by Mascott et al. who found no statistically significant correlation of RMS values with the accuracy of marker placement in a large in vivo study [
26]. Studies using phantom models typically measure the mean error of localization, which represents the average magnitude of the distance between the probe and its intended target. This should not be confused with the mean errors reported for individual axes in some in vivo [
27] and phantom [
28] studies which utilized planar imaging to measure targeting accuracy. In such instances mean errors refers to the average error within a single anatomic plane. A Euclidean error is then calculated as the square root of the sum of the squares of the mean errors in each dimension. Euclidean error is therefore generally larger than mean error, and more representative of the actual distance from a target one could reliably expect to achieve.
FB has traditionally been touted as being capable of sub-millimeter accuracy, though recent studies suggest that this may not be an entirely realistic expectation. Hall et al. [
29] found a euclidean error of 1.53 mm using a Leksell frame and MRI imaging in a phantom model. In a large phantom model study, Maciunas et al. [
30] found that while the mean mechanical errors of the CRW and BRW frames were less than 1 mm, at a 99.9% confidence interval they can only be expected to achieve a mechanical accuracy of 2 mm or less. When factors such as imaging, point selection, and vector calculations were all considered, the “application error” at the 99.9% confidence interval increased to 3.1–5.0 mm for four different frame systems. It is perhaps not practical, though, to make estimates of accuracy based upon an extreme limit of error that is likely to occur once in 1,000 cases. For this same series, the mean error of localization, or average distance between the probe and its intended target, was between 1.2 and 1.9 mm for the various systems tested. We feel that this is a more realistic estimate of the sort of accuracy one can expect in a given procedure. This corresponds well with in vivo assessments of FB for deep brain stimulator targeting accuracy, which found average stereotactic errors of 1.4–2 mm [
31].
There are several factors which likely contribute to the inherently higher accuracy of FB. These systems have their frame of reference rigidly fixed to the skull and established as soon as the frame is applied. Fiducial markers applied to skin are inherently more mobile and must be re-referenced to the navigation system once the patient is positioned, introducing two potential sources of error. Errors of imprecise trackable probe positioning as well as computer cursor positioning can be reduced, but not eliminated. The location of the fiducials has a significant impact on the zone of maximal correlation which increases accuracy and thus accuracy can vary considerably through various locations in a given registered target volume. The use of anatomic surface landmark registration in lieu of fiducial placement has consistently demonstrated lower accuracy rates [
32], particularly for posterior lesions where reliable anatomic landmarks are fewer.
In vivo assessments of FL systems have tended to yield larger error measurements than FB. Dorward et al. [
27] found a euclidean error of 4.8 mm in the in vivo arm of their study. Mascott et al. [
26] report mean localization errors between 3.3 and 5.4 mm. One relatively recent comparative phantom study actually found a smaller euclidean error in a frameless system when a specific planning and targeting protocol was used (probe’s eye) [
28]. It must be noted, as those authors themselves attest, that skull phantom models tend to overestimate the accuracy of FL systems, since such models simulate placing the fiducials directly upon the skull rather than the overlying skin, eliminating one major source of error in skin fiducial based systems. Indeed, in the Mascott study, mean localization error fell to 1.4–1.9 mm when skull impanted fiducials were used. Continued investigation in this realm is important as advances in computer image processing and algorithms enable neuronavigational systems to gain ground on the relatively established frame-based technologies. However, the above investigations indicate that currently, while it may be unrealistic to assume submillimetric accuracy for FB systems, it is probably reasonable to expect reliable targeting within 1–1.5 mm. Applying similar standards to FL systems using skin fiducials, that expected error rises to 3–4 mm and occasionally even greater depending on registration technique.
With this in mind, we devised our criteria of potential candidacy for FL biopsy. We found that fully 80% of our patients were candidates for frameless approach. Depending on which estimates of accuracy are used, FL systems can be expected to reliably target lesions of >5–10 mm. Grunert et al. [
33] came to a more conservative conclusion and suggested that lesions less than 15 mm should be reserved for FB approach. From a practical standpoint, requiring a lesion diameter of at least 10 mm rather than 5 mm excluded only two patients from the FL eligible group.
While based on accuracy considerations alone, >80% of biopsies could be accomplished through FL techniques, there are less easily quantifiable variables which ultimately influence technique selection. The risks of general anesthesia need to be considered, and the FB technique has traditionally been accomplished with local anesthesia and sedation. However in at least two comparative series, the surgeons induced general anesthesia in all patients [
8,
11]. This clearly influenced the outcome measure of total time and cost, which both studies found to favor FL. A comparative series conducted by Smith et al. [
10] which used local anesthetic for the FB arm found a substantial reduction in operating room time and cost with FB. While we utilized general anesthesia in 28.4% of cases, the remainder were done under local anesthesia. Frame-based stereotaxis has also typically been considered less invasive as it can usually be performed through a tiny stab incision and twist-drill craniostomy. While the FL technique could theoretically be performed through a twist drill craniostomy, rather than a burr hole, most FL technique series report the use of burr holes. Regardless, the three-point skull screw fixation, locking-ball-socket, device usually used for FL technique requires an actual incision to expose enough skull surface to seat the device and cannot be inserted through a simple stab wound. While 14% of our cases did require an incision and an open burr hole, this more invasive approach was safely avoided in the remaining 86%. Given the similarity of more objective measures such as diagnostic yield and complication rate between the two techniques [
10,
11], such discrepancies serve to highlight the important role individual surgeon training, preference and experience have in determining technique selection. For diagnosis of small deep-seated lesions, familiarity with frame-based techniques remains an important tool.