Draft:Stimulated Raman Histology (SRH)

• Comment: We can't accept AI / LLM submissions. ChrysGalley (talk) 13:07, 20 May 2026 (UTC)

Stimulated Raman histology (SRH) is a label-free optical image processing method that uses the spectral information obtained from stimulated Raman spectroscopy (SRS) to generate H&E stain like images. SRH utilizes the intrinsic vibrational signatures of the molecular bonds present in the biomolecules found in tissue samples to identify protein, nucleic acid, and lipid associated signals. Protein-rich regions can be computationally pseudo-colored in shades of pink to represent eosin staining, while nuclear and nucleic acid-rich regions can be pseudo-colored in shades of purple to represent hematoxylin staining. This allows for visualization and interpretation of tissue morphology using existing histological standards.

SRH is non-destructive and can be performed directly on un-processed tissue, with no need for fixation, sectioning, or chemical staining. Because of this, it is often used in cases where standard post-biopsy tissue processing is not possible, such as intraoperatively or when a non-destructive method is needed to perform multimodal analysis or preserve small sample volumes.

Principle

SRH relies on the Raman spectra obtained through SRS. In SRS, synchronized pump and Stokes laser beams are tuned so that their frequency difference matches a molecular vibrational mode, producing a unique signal depending on the molecular bonds that are being excited. A distinct Raman spectra is detectable at each point, depending on the combination of molecular components present within the sample.^[1] When SRS is being performed for the purposes of SRH, the Carbon-Hydrogen (C-H) stretching region of the Raman spectra is often targeted, as the signals from C-H bonds that are highly abundant in lipids and proteins are detectable in this region. While the C-H stretching region spans from 2,800–3,100 cm^-1 , papers using SRH commonly site using 2,840-2,850 cm⁻¹ for lipid detection and near 2,930-2,940 cm⁻¹ for protein detection. Upon imaging at the specified wavelengths, the grayscale maps obtained are then pseudo-colored to resemble H&E. This makes the output more familiar to pathologists to allow for standard diagnostic interpretation. Exact Raman bands and algorithms used are not standard across literature, but the main principle behind SRH remains the same. ^{[2][3][4][5][6][7]}

Clinical and research applications

SRH has been applied clinically as a method of intraoperative brain-tumor diagnosis, where it has been reported to allow for faster diagnosis with comparable accuracy within the operating room compared to conventional methods. In a multicenter prospective clinical trial, Hollon et al. reported that convolutional neural network (CNN) based diagnosis for major brain tumor classes was equivalent in accuracy to diagnosis by pathologists using conventional methods.^[3] A later central nervous system (CNS) telepathology clinical validation study reported high concordance between SRH-based diagnosis and conventional diagnosis, with a significantly shorter turn-around time.^[5] SRH workflows have also been demonstrated for use in skull-base tumors and gastroscopic biopsy diagnosis and analysis.^[4][6] Furthermore, workflows have been expanded to incorporate multimodal analysis. In a study by Fung et al., SRH was applied to diabetic kidney disease by developing a custom multi-modal optical-biopsy SRH pipeline that can also identify collagen morphology, mesangial-glomerular volume changes, lipid saturation, redox status, and protein/lipid features.^[7]

Advantages and limitations

SRH allows for label-free imaging, minimal sample preparation, short time to output, preservation of limited tissue, compatibility with telepathology, and straightforward integration with conventional and AI image analysis workflows. The method also allows for greater multimodal interpretation due to the use of molecular imaging data that captures more information than a standard H&E stain. Limitations to the application of SRH include the requirement for highly specialized instrumentation and the lack of available resources and training data for AI analysis models. In addition, the practice of reducing spectrally rich SRS data into H&E like images may lead to underuse of useful features in the underlying SRS channels. ^{[3][4][5][6][7]}

History

The development of SRH stemmed from the application of SRS microscopy to biomedical imaging in 2008.^[8] Early studies demonstrated the potential of SRS as a replacement to standard histology by distinguishing healthy from tumor-infiltrated brain tissue by interpreting differences in protein and lipid content using Raman spectra in 2013.^[2] However, it wasn't until 2017 that SRH was introduced formally as an image processing method by Orringer et al. in a study that used a portable fibre-laser SRS microscope with virtual H&E rendering to demonstrate the potential of SRH for use in intraoperative diagnostic procedures.^[9] Later works expanded the approach to applications including: automated brain-tumor classification, multimodal kidney pathology, telepathology validation, skull-base tumors, and rapid gastroscopic biopsy assessment.^{[3][7][4][5][6]}

References