Intraoperative tissue identification

The classic workflow carries a structural flaw: the surgeon removes tissue, the pathologist examines it afterwards, and any discrepancy becomes a second operation. Frozen-section pathology compressed the wait to twenty or thirty minutes at the cost of interrupting anaesthesia and sampling only slivers of the margin. The intraoperative toolset now emerging answers the same question — tumour or healthy? — by different physical routes, each trading preparation time against coverage and specificity.

Making tumours glow on command

Targeted fluorescence exploits metabolic or receptor biology so that malignant tissue labels itself. Gliomas, rewiring their heme synthesis, import 5-aminolevulinic acid and accumulate fluorescent protoporphyrin IX; under blue-light illumination the tumour glows violet against quiescent brain, drawing resection boundaries that white light cannot resolve. Epithelial cancers overexpressing the folate receptor likewise concentrate an injected folate-dye conjugate, letting ovarian and lung tumours announce themselves during surgery. The mechanism’s elegance — selective accumulation does the analytical work, optics merely reads it — carries its own caveat: selectivity is relative, since inflamed tissue expresses the same receptors and non-enhancing tumour regions stay dark. Fluorescence guides toward; it does not certify.

Tissue identity from its physics

Label-free approaches interrogate chemistry directly. Stimulated Raman histology reads the vibrational spectra of lipids and proteins — every tissue class holds a distinctive fingerprint — then hands those spectra to a network trained to render them as synthetic images resembling conventional stained histology, deliverable within minutes and classified alongside by algorithms built for digital pathology. Surgical smoke offers a stranger sampling geometry: electrosurgical cutting aerosolises cellular material, and rapid mass-spectrometric analysis of that plume identifies tissue type second by second as the blade moves — the diagnostic arrives attached to the act of cutting. Radiofrequency spectroscopy takes a third route, sensing dielectric differences between tumour and breast tissue through a contact probe. All three share a philosophy: skip the stains, skip the transport, classify matter as encountered.

Latency and coverage decide, not resolution

Two constraints organise the field more than any technique’s sophistication. First, decision latency: every intraoperative minute is an anaesthetised minute, so any answer arriving after half an hour reshapes surgery less than one arriving in ninety seconds; this is why optical methods crowd out laboratory workflows at the table. Second, sampling fraction: frozen section examines a few selected slivers — the margin mostly goes unread — whereas optical scanning interrogates an entire excised surface continuously, trading a little per-point certainty for the elimination of unsampled territory. Which trade favours the patient depends on how dangerous residual disease is versus how disruptive false alarms prove; breast-conserving surgery, where margins drive re-excision rates, currently argues for full-surface coverage.

Honest limits trail every modality: optical penetration stops within millimetres of the surface, fluorescence-based methods inherit their tracer’s biology including its false positives, and each algorithm has been validated against final pathology in defined tumour types — definitions that do not yet transfer freely between cancers or scanners. The direction, though, is structural: diagnosis migrating from the laboratory back into the operating room, completing while the wound is still open.