Scientists map plan to test Raman spectroscopy for real-time glioma diagnosis

A team of researchers has published a detailed protocol for a systematic review that will evaluate whether Raman spectroscopy, a label-free optical technique, can serve as a reliable diagnostic tool for glioma, the most common primary brain tumor. The protocol, registered with PROSPERO under identifier CRD420251025922, lays out a rigorous plan to synthesize existing evidence on the technique’s ability to detect glioma micro-infiltration, particularly in the operating room where decisions about how much tumor to remove must be made in minutes. The work appears in the journal Diagnostic and Prognostic Research and follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses, or PRISMA, 2020 guidelines along with the PRISMA-P checklist designed specifically for review protocols.

Glioma carries an annual age-adjusted incidence rate of 6.0 per 100,000 population, and the standard treatment for malignant glioma is maximal safe surgical resection. The challenge facing neurosurgeons is that gliomas invade surrounding brain tissue in ways that resist precise visual delineation. Residual malignant cells left behind after surgery often drive rapid postoperative recurrence, yet removing too much tissue risks permanent functional impairment. The protocol’s authors note that determining tumor boundaries and striking an equilibrium between the harm of excessive resection and the risk of progression with insufficient resection is essential in glioma treatment. Current imaging modalities, including magnetic resonance imaging, fluorescence-guided imaging, and positron emission tomography, provide valuable intraoperative information, but their resolution is insufficient to detect micro-infiltrative tumor cells, making visualization of glioma margins a persistent challenge.

The intraoperative diagnostic gold standard remains histopathological assessment, but the protocol points out that combining histological and molecular features of glioma is time-consuming and limited in spatial coverage, raising costs for patients. Microscopic diagnosis requires the intervention of multiple experienced pathologists analyzing hematoxylin and eosin-stained slides. Invasive surgical resection or biopsy of glioma tissue, followed by Sanger sequencing and fluorescence in situ hybridization, further complicates the diagnostic pathway. Raman spectroscopy offers a fundamentally different approach. Rather than relying on stains or dyes, the technique detects the unique molecular fingerprint of tissues, enabling real-time differentiation between tumor-infiltrated and normal brain based on biochemical composition. Recent studies have used Raman spectroscopy to demonstrate these differences at the cellular level, and the method has shown considerable promise in identifying biochemical changes associated with early-stage glioblastoma infiltration.

Advances in laser technologies, fiber optics, and machine learning have further facilitated the integration of Raman spectroscopy into clinical workflows, enhancing its potential as a real-time diagnostic tool. The technique has shown promise for identifying glioma in real time, particularly when combined with machine learning algorithms for spectral classification. Despite this growing body of experimental and translational work, the authors identify a critical gap: no comprehensive synthesis of Raman spectroscopy’s clinical applicability in detecting glioma micro-infiltration exists. The new protocol is designed to fill that void by critically evaluating current evidence on the diagnostic utility of Raman spectroscopy in glioma, with a focus on intraoperative performance, methodological consistency, and translational readiness.

The review will follow a structured Patients, Intervention, Comparison and Outcomes framework. Eligible studies must involve adult patients aged 18 years or older with suspected or histologically confirmed glioma undergoing diagnostic evaluation, surgical intervention, or guided-tissue sampling using Raman spectroscopy. The intervention of interest is Raman spectroscopy used either alone or combined with machine learning algorithms as a diagnostic approach for detecting and characterizing glioma. Comparators include standard diagnostic modalities such as intraoperative frozen section analysis, confirmed postoperative histopathology, magnetic resonance imaging, fluorescence imaging, and positron emission tomography. Studies involving pediatric populations, non-human subjects, or glioma cell lines will be excluded, as will studies where the patient population is not clearly described or where glioma-specific data cannot be extracted.

The search strategy is comprehensive and transparent. Researchers will query four electronic databases: PubMed, Embase, Scopus, and Web of Science. The principal keywords are Raman and glioma, with the Raman term expanded to include synonyms and variations such as Raman spectroscopy, Raman spectrum, Raman scattering, Raman imaging, Raman optical activity, and Raman spectrometer. The glioma term will likewise be expanded to cover glioblastoma, glioblastoma multiforme, high-grade glioma, anaplastic astrocytoma, diffuse glioma, recurrent glioma, glial tumor, brain neoplasm, and central nervous system tumor. These two sets of terms will be combined using Boolean operators to ensure results reflect studies discussing both Raman spectroscopy and glioma. The search will include English-language studies published between January 2015 and the date of protocol registration, May 30, 2025, with a final search update conducted before data extraction to capture the most recent studies. Database searches will be split among team members, with specific reviewers assigned to specific databases, and all results will be combined and deduplicated in a reference manager before screening.

Study selection will involve two independent reviewers performing both title/abstract screening and full-text screening according to predefined eligibility criteria. Discrepancies will be resolved by involving a third reviewer, and the number of discrepancies at both screening stages will be recorded. Inter-reviewer agreement will be reported using Cohen’s kappa coefficient. Full texts of potentially eligible studies will be assessed in detail, with exclusions documented systematically for reasons such as not using Raman spectroscopy, not involving glioma, not being original research, or lacking diagnostic data. Preprints without peer review and retracted articles will be excluded. The review team will verify publication status and screen for retractions before data synthesis. A PRISMA 2020 flow diagram will document the entire selection process.

Primary outcomes for the review will be diagnostic accuracy measures, including sensitivity, specificity, accuracy, area under the curve, positive predictive value, negative predictive value, diagnostic odds ratio, and positive and negative likelihood ratios. Secondary outcomes will explore the technique’s role in surgical margin assessment and intraoperative guidance, and, if data are available, speed of intraoperative decision-making, the contribution of machine learning techniques to diagnostic performance, feasibility of application in clinical settings, and diagnostic performance differences according to sample type or tissue location. Two reviewers will independently extract data using a standardized Excel sheet capturing study characteristics, population details, Raman spectroscopy details such as type, wavelength, and analysis method, comparator used, and diagnostic outcomes. The extraction will also capture machine learning or statistical classifiers applied for processing, feature extraction, dimensionality reduction, and classification of Raman spectral data, including model types such as principal component analysis, support vector machine, and convolutional neural networks.

Risk of bias for all included diagnostic accuracy studies will be evaluated using the QUADAS-2 tool, which assesses internal validity across four domains: patient selection, index test, reference standard, and flow and timing. Each domain will be assessed for risk of bias and applicability concerns. If 10 or more studies provide proportionate diagnostic values such as sensitivity and specificity, a meta-analysis will be conducted using a bivariate random-effects model to generate pooled estimates including sensitivity, specificity, diagnostic odds ratio, and the area under the summary receiver operating characteristic curve, each with 95 percent confidence intervals. Statistical heterogeneity will be assessed using the I2 statistic alongside visual inspection of SROC curves and consideration of potential threshold effects. Subgroup analyses will be performed based on Raman modality, application context, and whether machine learning was used. The potential influence of machine learning methodology, including preprocessing strategies and validation approaches, as well as sources of bias such as overfitting, lack of external validation, and data leakage, will be critically examined as contributors to heterogeneity. If meta-analysis is not feasible, a structured narrative synthesis will be employed following the Synthesis Without Meta-analysis guidelines.

The protocol also addresses publication bias and selective outcome reporting through a structured approach adapted to the number of studies available. When at least ten studies are included for a given outcome, publication bias will be evaluated both visually and statistically using funnel plots, Egger’s regression test, trim and fill, and small study effects. When fewer than 10 studies are available, a narrative assessment will explore potential sources of bias, examining factors such as study sample size, precision of estimates, funding sources, publication status, and consistency of outcome reporting relative to study protocols or trial registrations. The GRADE approach will be used to assess the certainty of evidence across studies for each outcome, and a Summary of Findings table will be created using GRADEpro GDT. The authors acknowledge that heterogeneity of Raman systems, analysis methods, and sample types may introduce substantial variability in diagnostic accuracy estimates and affect comparability across studies. They also note that the clinical value of Raman spectroscopy will depend not only on diagnostic accuracy but also on usability, reproducibility, and cost-effectiveness, all of which are underreported in primary studies. The review aims to map current evidence gaps, including the lack of prospective multicenter trials, absence of standardization in Raman data acquisition and preprocessing, and limited regulatory pathways, ultimately outlining a translational roadmap for integrating real-time Raman spectroscopy into neurosurgical workflows with machine learning support.

Subject of Research: Systematic review protocol evaluating Raman spectroscopy for glioma diagnosis

Article Title: Raman spectroscopy as a diagnostic tool for glioma: protocol for a systematic review

Article References: Vrancianu, C. O., Necula, M., Milea, E., Siminea, N., Cristian, R.-E., Stoica, A., Chifiriuc, M. C., Oancea, B., Paun, A., Petre, I., & Paun, M. (2026). Raman spectroscopy as a diagnostic tool for glioma: protocol for a systematic review. Diagnostic and Prognostic Research, 10(1), Article 9. https://doi.org/10.1186/s41512-026-00226-x

Image Credits: AI Generated

DOI: 10.1186/s41512-026-00226-x

Keywords: Raman spectroscopy, glioma, systematic review protocol, diagnostic accuracy, intraoperative margin assessment, machine learning, PRISMA, QUADAS-2, GRADE, neurosurgical oncology, PROSPERO, glioblastoma