Intraoperative imaging

For almost the whole history of neurosurgery, the surgeon learned whether the tumour had been fully removed only the next day, when the control MRI arrived. If something was left, it was already too late: reoperating meant a second anaesthetic, a second opening and a second risk.

Intraoperative imaging removes that wait. The field is temporarily closed, the study is performed with the patient still asleep and, if tumour remains and it is safe to remove, the surgeon goes back in. It is a good and obvious idea, which is precisely why it is instructive to look at what happened when it was measured.

What the foundational trial shows

In 2011 the first randomised trial of intraoperative MRI in glioma surgery was published¹. Fifty-eight patients randomised; 49 analysable. Complete resection was achieved in 96 per cent of patients operated with MRI and 68 per cent of the control group. New neurological deficits: 13 against 8 per cent, not significantly different. And a sentence its authors emphasise: none of the patients in whom the image led to continued resection deteriorated neurologically.

It is a solid result in what it measures. What it did not measure is survival: the primary endpoint was the complete resection rate, the study was neither designed nor powered to detect differences in how long patients live, and the analysis was per protocol rather than by intention to treat.

The Cochrane review that assessed all randomised evidence assigns intraoperative MRI very low certainty, based on 49 patients from a single trial². And the network meta-analysis that set out to compare technologies could not be performed, because of heterogeneity and risk of bias across all included studies.

The two trials that came later

In 2024 the largest trial in the field was published: 321 randomised patients³. Total resection rose from 50.00 to 83.85 per cent with intraoperative MRI, an enormous and highly significant difference. In high-grade gliomas overall survival was 29.73 against 25.33 months, with a p value of 0.1233: not significant. In low-grade tumours it was not significant either, in survival or in time to progression.

And that same trial reports a finding the observational literature had not detected: grade 3 intracranial infection in 18.01 per cent of patients with intraoperative MRI against 9.38 per cent of controls, statistically significant. A later observational study of 446 patients found no difference in surgical site infection, but with 114 patients in the MRI arm it lacked power to rule out an effect of that size. It is the most interesting pair of data points in the field, and until another trial confirms or refutes it, prudence says to treat it as a real signal.

The second uncomfortable result arrived in 2023, from eleven German centres and 314 patients with glioblastoma⁴. The comparison was not against conventional surgery but against 5-ALA fluorescence. Complete resections: 81 per cent with MRI, 78 with fluorescence, no difference. Overall and progression-free survival, comparable. Incision-to-suture time: 316 minutes with MRI against 215 with fluorescence. A hundred extra minutes of theatre for the same result. The authors' conclusion is literal: we could not confirm superiority of intraoperative MRI over 5-ALA.

A network meta-analysis that admitted observational studies found MRI to quintuple the odds of total resection against conventional navigation, and that the indirect comparison with 5-ALA does not reach significance⁵. The contrast between that synthesis and Cochrane's is itself instructive: the same body of literature yields a striking effect if cohorts are admitted and "very low certainty" if randomised trials are required.

What it costs and how long it takes

These are the figures that almost never appear in presentations.

A cost analysis of a low-field intraoperative MRI theatre, across 196 patients, puts acquisition and installation of the equipment at one million euros, and the amortised cost per procedure at 833 euros assuming a ten-year life and 120 procedures a year⁶. The MRI lengthened surgery by 47 minutes on average: 415 against 368. And the conclusion about benefit is characteristically qualified: better functional outcome, more resection, fewer complications, better progression-free time and a similar life expectancy.

A US cost-utility model estimated an incremental cost of 13,447 dollars per patient and a ratio of 76,442 dollars per quality-adjusted life year, below the usual willingness-to-pay threshold⁷. It is a model, not a measurement, and it depends critically on assuming a quality-of-life gain derived from observational studies.

And there is the real procedural time, which is not the time of the scan. A series that solved the problem without building a dedicated theatre — transferring the patient to the radiology department with the wound temporarily closed — measured a median of 68 minutes from the decision to scan to reopening⁸. It detected residual tumour in 54 per cent of cases, with no complications or infections. It is an elegant and far cheaper solution; it also makes explicit what the image really costs in minutes.

The finding nobody expected

There is a question prior to all the above, and almost nobody had asked it: when the MRI shows residual tumour, does the surgeon go back in?

A series of 486 patients across seventeen years answered it⁹. After residual tumour was visualised, further resection was performed in 47 per cent of high-grade gliomas, 45 per cent of low-grade gliomas, 29 per cent of pituitary adenomas and in none of the meningiomas or metastases. In more than half of gliomas with visible residual, the team chose not to continue.

That is not a failure of the technology: it may be sound judgement working, if what remained sat somewhere untouchable. But it changes the nature of the tool. Intraoperative MRI does not increase resection on its own; it increases resection in the fraction of cases where the surgeon was already willing to go further. In that same work, the overall further-resection rate rose from 33 per cent in the first five-year period to 46 in the last, suggesting that learning to use the image takes years and is an institutional rather than a technical process.

The cheap alternative

Intraoperative ultrasound costs two orders of magnitude less, requires no building works, does not move the patient and gives real-time images. Its problem is that it is harder to read.

One study compared linear-array ultrasound, conventional sector-array ultrasound and high-field intraoperative MRI against the correct reference standard, which is the histology of 68 navigated biopsies¹⁰. Sensitivity was 76 per cent for linear ultrasound, 55 for MRI and 24 for sector ultrasound. The specificities run the other way: 58, 74 and 96 per cent. The counterintuitive datum is the MRI's: against histology, it missed almost half of the histologically confirmed residual tumour. No image sees the infiltrating margin.

A meta-analysis of thirteen studies and 665 gliomas gives ultrasound a pooled sensitivity of 72.2 per cent and specificity of 93.5¹¹. And the only randomised trial of ultrasound repeats the familiar pattern: total resection rose from 8 to 35 per cent, and overall survival was 377 against 372 days, with no difference at all¹².

There is also an artefact worth knowing because it is treacherous. The fluid filling the resection cavity does not attenuate ultrasound the way brain does, so the tissue beneath appears abnormally bright. That enhancement can both conceal a tumour remnant and cause healthy tissue to be resected in the belief that it is tumour¹³. The way to tell them apart is to compare successive acquisitions: what was bright from the start is probably tumour; what appeared only once the cavity filled with saline, probably is not.

A result that inverts the intuition

In spine surgery intraoperative imaging means CT, and the usual argument is pedicle screw accuracy. A meta-analysis of thirteen studies in paediatric deformity found that screws placed with CT navigation are three times as likely to be judged acceptable and one third as likely to be potentially unsafe¹⁴. Operating time was about thirty minutes longer.

But the radiation dose did not fall. In that meta-analysis the difference did not reach significance, and it trended towards more dose with navigation. A study that measured exposure directly found a patient dose of 43.2 millisievert with CT navigation against 27.7 with fluoroscopy, while the dose to the surgeon's hand fell from 566 to 49 microsievert¹⁵. Intraoperative CT does not reduce radiation: it redistributes it. It shifts the dose from the surgical team to the patient, who is also the one exposed fewest times in a lifetime. It is a defensible trade and it should be made with open eyes.

What to take away

Intraoperative imaging is the technology that best demonstrates the gap between the outcome that gets measured and the one that matters. It reproducibly increases the proportion of complete resections — that is solid, across three randomised trials — and in none of them did survival improve significantly.

Perhaps the trials are too small. Perhaps complete resection measured by contrast enhancement is not the right endpoint, because glioma recurs precisely where it does not enhance. Perhaps the benefit sits in subgroups that were not identified. All those explanations are reasonable.

What is not reasonable is presenting extent of resection as though it were survival. They are two different claims, and the second still lacks support.