Neuronavigation and tractography
The navigation system tells you where the tumour is to within millimetres. The problem is that the brain, the moment the dura is opened, stops being where the image says it is.
Neuronavigation does something that sounds elementary and is not: it takes the MRI acquired days before surgery and superimposes it on the actual skull in the operating room, so that the tip of a pointer in the surgeon's hand appears as a dot moving inside that image. It is, literally, a GPS. And like any GPS, it works well as long as the territory does not change.
Tractography goes one step further. Instead of showing the tumour, it draws the bundles of nerve fibres around it: the connections that carry movement, language, vision. It shows the motorways that must be preserved. It is the most eloquent image in modern neurosurgery — the one used in lectures and brochures — and it is also the most misread.
That distinction is worth taking seriously, because the correct use of both tools depends on it.
What the only randomised trial says
In 2006 a Dutch group published what remains, two decades later, the only randomised trial of neuronavigation in brain tumour surgery1. Forty-five patients, randomly assigned to surgery with or without the navigation system. The result was not the expected one: residual tumour volume did not differ significantly between groups, and gross total resection was achieved in five patients without navigation and in three with it. The authors' conclusion is explicit: there is no rationale for routine use of neuronavigation to improve extent of resection when the lesion does not already demand it by its size or location.
The trial also found shorter survival in the navigation group. That figure must be read carefully: it is an unadjusted hazard ratio, in 45 patients, and it almost certainly reflects that the harder cases ended up in the navigated arm. It does not say the navigation system does harm. It says the study did not find the benefit everyone assumed.
The most revealing datum is not in that trial but in the systematic review that assessed it. When the Cochrane group searched for all randomised evidence on intraoperative imaging technologies in glioma, it found four trials in total, and the only one of neuronavigation was that one, with its 45 participants2. All of contemporary neuronavigation — present in any decent operating room in the world — rests, in terms of randomised evidence, on a single small negative study.
This does not mean the navigation system is useless. It means its usefulness is of a different order: planning the approach, choosing the size and position of the craniotomy, orienting oneself in distorted anatomy. Those are real benefits and hard to capture with a hard endpoint. But it is worth saying out loud that they were never demonstrated.
Why the map goes out of date
There is a physical reason why navigation cannot do what its metaphor promises. A car GPS works because streets do not move. The brain does.
Start with the error at the outset. A study measuring the accuracy of two navigation systems in 55 consecutive patients found that, as soon as initial registration was complete, the mean error was already 2.9 millimetres3. And from there it only worsens, through gestures nobody would associate with precision: draping adds 2.7 millimetres; attaching the retractor, another millimetre; the craniotomy, another. With time the degradation is relentless: 1.3 millimetres of drift at half an hour, 4.4 at five and a half hours. The authors call it, accurately, "the silent loss of accuracy".
Then there is brain shift proper. When the dura is opened, intracranial pressure is released and the brain moves. A classic 1998 paper measured a mean cortical surface displacement of about a centimetre, with direction dominated by gravity4. A later review compiles the published maxima: up to 10 or 13 millimetres from opening the dura alone, before resecting anything, and up to some 24 millimetres after emptying the cavity5. The midline, by contrast, barely moves. Which is to say: the deformation is not uniform, and therefore cannot be corrected by a global shift.
And the tracts move with everything else. A study of 37 patients measured white matter displacement during surgery and found a range of −8 to +15 millimetres6. The most uncomfortable part of the finding is the absence of direction: in about a third of cases the tract sank inward and in nearly two thirds it bulged outward. No simple correction is possible when the error has no predictable sign.
Tractography, measured against the truth
How well does tractography draw what it claims to draw? The question has an answer, and it is not comfortable.
The reference standard in brain surgery is not another image: it is direct electrical stimulation. Current is applied at a point in the tissue and one observes whether the patient's hand moves. If it moves, there is motor tract there. When the two methods are compared point by point, the mean distance between stimulation sites that gave a response and the tract drawn by tractography was 8.7 millimetres, with a standard deviation of 3.17. Almost a centimetre between the real tract and the one on the screen.
A more recent study quantified it differently: for tractography based on the diffusion tensor — the classic method, the one most commercial software ships with — to reach acceptable sensitivity against stimulation, one must accept a margin of 14.5 millimetres. With more modern spherical deconvolution algorithms the margin falls to 8.58. In surgery for a glioma in eloquent territory, a margin of nearly a centimetre and a half is enormous.
The underlying problem was demonstrated by a remarkable collective experiment. Twenty research groups from twelve countries submitted 96 different tractography pipelines applied to a simulated brain whose correct answer was known. The algorithms recovered on average 21 of the 25 true bundles — good sensitivity — but they also fabricated 88 bundles that did not exist. Mean precision at the bundle level was 23 per cent9. Put differently: of every four bundles an algorithm draws, roughly three are false. And this is not random noise: 41 of those nonexistent bundles appeared systematically across most submissions.
The authors themselves warn, with a candour worth imitating, that those proportions come from a simulation and should not be transferred literally to a patient. But the result held when the exercise was repeated against real histology, with tracers injected in primates: 176 submissions from nine groups, and the conclusion that the anatomical accuracy of tractography has not improved substantially in recent years10.
So what is it good for?
It is good for a great deal, if used for what it is.
A prospective trial of 238 patients with gliomas near the pyramidal tract compared operating with and without functional navigation based on tractography. In high-grade tumours gross total resection was 74.4 per cent with tractography against 33.3 without it, and postoperative motor deterioration fell from 32.8 to 15.3 per cent11. These are large differences, from a single centre and with no possibility of blinding the surgeon, but they point in a clear direction.
What tractography does not achieve is replacing stimulation. The meta-analysis comparing both strategies across 1,837 patients found more complete resection with tractography, but no difference in deficits: 35.45 against 35.60 per cent for early ones, 6.00 against 4.91 for late ones12. And that comparison is misleading by construction, because patients who undergo stimulation mapping are, by definition, those with the most eloquent tumours.
The most recent systematic review of intraoperative tractography gives the exact measure of the state of the evidence: it pools a complete resection rate of 79 per cent with between-study heterogeneity of 93.4 per cent13. A value like that means the studies are so different from one another that the average is nearly meaningless. It is also worth noting that two of that review's authors have a patent pending on tractography technology, and they declare it. That does not invalidate the work; it makes it legible.
What to take away
Hugues Duffau, the surgeon who has most forcefully defended awake mapping, titled a commentary "The Dangers of Magnetic Resonance Imaging Diffusion Tensor Tractography in Brain Surgery"14. The title is a position, and it is the right one.
A tractogram is not a finding: it is an anatomical hypothesis, generated by an algorithm, from an image taken days earlier, in a brain that has since moved more than a centimetre. Used as a hypothesis — where to enter, what to expect, where to pay closer attention — it is a valuable tool. Used as a certainty, it is an elegant way of being confidently wrong.
The practical rule that follows from all of the above is simple: navigation orients, stimulation decides. And when the two disagree, the one that is right is the one plugged into the patient.