Detection of Human Brain Tumor Infiltration With Quantitative Stimulated Raman Scattering Microscopy

Better imaging methods to accurately resolve glioma margins may help improve resection and clinical outcomes for glioma patients. Currently, clinical imaging technologies cannot reliably visualize infiltrating glioma cells, leading to incomplete resections and tumor recurrence. Recently, Ji et al1 demonstrated improvements in the use of stimulated Raman scattering (SRS) microscopy in a label-free, automated fashion to accurately delineate tumor margins in ex vivo tissue specimens. SRS microscopy relies on differences in the intrinsic vibrational properties of lipids, proteins, and DNA to achieve chemical contrast. It does not require labeling and can be performed in situ. The different compositions of these macromolecules in malignant and normal tissue can be detected with SRS microscopy to distinguish malignant tissue from normal tissue at the cellular level. A dual Raman frequency approach measuring the ratio of Raman signals at 2930 and 2845 cm−1 (S2930/S2845) reflects the different protein and lipid concentrations of brain regions, with highly cellular regions appearing more protein dense and areas of dense axonal regions appearing more lipid dense. As initial proof of principle, the authors demonstrated that SRS imaging using the protein channel (2930 cm−1) and lipid channel (2845 cm−1) recapitulates many histological features of normal brain specimens and histopathological hallmarks of different central nervous system malignancies. When neuropathologists were shown images of SRS microscopy–analyzed biopsy specimens and hematoxylin and eosin–stained tissue histology from 3 control epilepsy patient brains, 2 low-grade gliomas, and 2 high-grade gliomas, SRS analysis accurately distinguished between normal brain, infiltrating glioma, and high-density glioma with similar accuracy (95.1% vs 92.4%, respectively). To further minimize analysis time, the authors automated the process of determining tumor infiltration vs no tumor infiltration by quantifying salient features such as nuclear density, axonal density, and protein/lipid ratio (Figure, A). The program was able to accurately and automatically quantify these measures with very similar results compared with manual quantification using a set of 1477 fields of view (FOVs) obtained from 51 fresh tissue biopsies of 18 patients (3 epilepsy control subjects and 15 patients with brain cancers). The authors then derived a classifier system using half the FOVs by integrating all 3 metrics into a single probability score to distinguish tumor infiltration from no infiltration (Figure, A). The classifier system detected tumor infiltration with 97.5% sensitivity and 98.5% specificity in the other half of the FOVs. This classifier system was also highly accurate in distinguishing between different categories of tumor infiltration: normal, infiltrating glioma, and dense glioma (Figure, B). Because glial tumors tend to have less distinct borders than nonglial tumors, a separate classification system was developed for detecting glial tumor infiltration using the same metrics, leading to 97.0% sensitivity and 98.5% specificity. Finally, because these models incorporated FOVs from the same patients to both derive and test the classifier systems, a separate classifier system was developed that excluded a patient from the derivation set to eliminate potential dependencies. This “leave-one-out” cross-validation system predicted tumor infiltration in the excluded patient with 87.3 sensitivity and 87.5% specificity.Figure: Nuclear density, axonal density, and ratio of protein to lipid are quantified from stimulated Raman scattering (SRS) images to derive classifier values. A, 1477 fields of view (FOVs; 300 × 300 mm2) from 51 fresh tissue biopsies from 18 patients (3 epilepsy patients and 15 patients with brain and spine tumors encompassing 8 distinct histological subtypes) were quantified for nuclear density, axonal density, and ratio of protein to lipid on the basis of SRS microscopy analysis. Each point on the scatterplot represents the average value of each biopsy, and each biopsy was classified as predominantly normal to minimally hypercellular (n = 21), infiltrating tumor (n = 14), or high-density tumor (n = 16) by a board-certified neuropathologist on the basis of hematoxylin and eosin staining. Marker color indicates the mean classifier value for each biopsy, with 0 (most likely normal) depicted in cyan and 1 (most likely tumor) depicted in red. Representative FOVs from normal cortex, normal white matter, low-grade glioma, and high-grade glioma are shown. Green represents lipid-dense areas (S2930/S2845 >1); blue represents protein-dense areas (S2930/S2845 <1). B and C, relationship of classifier values with tumor density (B) and histological subtype (C). All parameters are normalized to the maximum measurement obtained of that variable and displayed in arbitrary units. Data are mean ± SEM. GBM, glioblastoma multiforme. Modified from Ji et al. From Ji M, Lewis S, Camelo-Piragua S, et al. Detection of human brain tumor infiltration with quantitative stimulated Raman scattering microscopy. Sci Trans Med. 2015;7(309):309ra163. Reprinted with permission from AAAS.SRS microscopy is a sensitive method to detect glioma margins and histopathological hallmarks of central nervous system malignancies without laborious labeling or processing of biopsied specimens. Ji et al1 have further refined SRS microscopy to an automated, quantitative approach that may be more easily integrated into clinical workflow to detect infiltrating gliomas with accuracy. Further work using larger, independent data sets will improve the sensitivity and specificity of the automated classifier system. Although this method cannot provide all the architectural, genetic, and biochemical data of traditional molecular and histological analysis, it can potentially be useful intraoperatively to determine the glioma margins in situ or ex vivo to improve resections. Evaluating in situ SRS microscopy and exploring strategies to coregister the SRS imaging data (currently limited by depth) with the surgical FOV are underway to further realize the clinical potential for SRS microscopy.