Hyperspectral Imaging Tool Maps Brain Tumor Metabolism Without Dyes or Stains

Gliomas are hard to treat because they infiltrate surrounding brain tissue and often recur. Surgeons plan with MRI, CT, and PET, but brain shift during surgery makes these images less reliable. Fluorescence-guided surgery with 5-ALA or fluorescein helps, but it depends on injected dyes and has limited specificity. Final tumor grading comes from histopathology, which is subjective and can take weeks. A fast, label-free, spatially resolved tool could help at both stages.

The researchers developed HyperProbe1.1 (HP1.1), a hyperspectral imaging system built around a broadband plasma light source and a tunable wavelength selector. It scans 385 to 1015 nm in 5 nm steps and builds a hyperspectral cube in about 30 seconds. A separate 370-nm LED module excites NADH and FAD autofluorescence, which is read through 450/20 and 520/20 nm filters. A Beer-Lambert-based spectral unmixing algorithm run over 500 to 900 nm estimates oxy- and deoxy-hemoglobin, lipids, water, cytochrome c oxidase (oxidized and reduced), cytochrome b, and cytochrome c. The team validated the system on MEDPHOT plastic phantoms with known absorption and scattering, and on fluorescein and protoporphyrin IX liquid phantoms. They then imaged fresh biopsies from 31 enrolled patients with suspected glioma, of whom 28 were analyzed (1 grade 1, 5 grade 2, 3 grade 3, 19 grade 4). Each specimen was split, and one half underwent H&E and IDH1 R132H histopathology under the 2021 WHO classification as ground truth. Imaging occurred within 1 hour of excision. Sixteen samples also underwent NADH/FAD autofluorescence imaging.

On phantoms, HP1.1 showed high contrast on a USAF target and resolved group 6 element 5 over a 6.5 mm field of view. Signal-to-background ratio rose linearly with exposure time from 10 to 100 ms. Reflected intensity from a brain-mimicking disk stayed stable over hours, days, and weeks. Attenuation correlated linearly with absorption, scattering, and their ratio, indicating quantitative behavior across optical regimes. The system also detected fluorescein attenuation at low concentrations. In glioma samples, images at different wavelengths showed different structures. Shorter wavelengths gave finer detail, and near-infrared gave deeper but more diffuse views. The authors report that the molecular maps distinguished glioma tissues of different histological grades. The provided text is truncated, so detailed grade-separation statistics are not available.

If validated further, this approach could give surgeons rapid maps of vascularization, oxygenation, and metabolism without dyes, and it could support early postoperative grading while histopathology is pending.

Several caveats apply. The work is a laboratory study on ex vivo samples. The cohort is small and skewed toward grade 4, with only one grade 1 sample. No healthy or margin tissue was imaged, for ethical reasons, so tumor-versus-normal discrimination was not tested. One histological label was applied to a whole specimen despite tumor heterogeneity. The 30-second acquisition and 2 to 3 minute unmixing time also fall short of true real-time intraoperative use.