Stimulated Raman histology for histological evaluation of oral squamous cell carcinoma

David Steybe, Philipp Poxleitner, Marc Christian Metzger, René Rothweiler, Jürgen Beck, Jakob Straehle, Kirstin Vach, Andreas Weber, Kathrin Enderle-Ammour, Martin Werner, Rainer Schmelzeisen, Peter BronsertClinical 2023 · DOI

TL;DR

This paper explores Stimulated Raman Histology (SRH) as a novel, label-free imaging technique for intraoperative evaluation of oral squamous cell carcinoma (OSCC). SRH shows high accuracy in distinguishing cancerous from healthy tissue and can also aid in classifying non-cancerous tissues, offering a promising alternative to traditional, time-consuming H&E staining of frozen sections.

What Problem Does This Paper Address?

Oral squamous cell carcinoma (OSCC) is a prevalent and aggressive cancer, with surgical resection being a cornerstone of its treatment. During surgery, accurately differentiating between cancerous and healthy tissue is crucial for ensuring complete tumor removal while preserving vital structures to maintain function and quality of life. The current standard for this intraoperative assessment relies on Hematoxylin and Eosin (H&E) stained frozen sections. However, this method is labor-intensive, requires specialized laboratory infrastructure and trained personnel, and involves time-consuming processing steps. This study investigates Stimulated Raman Histology (SRH) as a potential solution to overcome these limitations by providing rapid, label-free, digital histopathology images of fresh tissue specimens without the need for sectioning or staining.

Methods

The researchers employed Stimulated Raman Scattering (SRS) microscopy, specifically using the NIO Laser Imaging System. This technology utilizes Raman scattering, a phenomenon where photons interact with molecules and scatter at different frequencies, to probe the chemical composition of tissues. The system was configured to detect Raman shifts corresponding to CH2 bonds (abundant in lipids) and CH3 bonds (abundant in proteins and DNA).

For imaging, fresh tissue samples were placed on a custom microscope slide and gently compressed. The NIO Laser Imaging System then acquired images by performing multiple line scans. These scans were automatically stitched to create virtual sections approximately 4-5 µm thick. The raw data was then processed by the system's software to generate "pseudo H&E images," which are designed to visually resemble conventional H&E-stained slides. This process is referred to as Stimulated Raman Histology (SRH).

For comparison, conventional H&E-stained frozen sections were prepared from the same tissue samples. Both SRH images and H&E sections were then analyzed by pathologists and surgeons to identify various tissue types, including OSCC, normal mucosa, connective tissue, muscle tissue, adipose tissue, salivary gland tissue, lymphatic tissue, and inflammatory cells.

Datasets

The study utilized 80 tissue samples obtained from 8 patients diagnosed with oral squamous cell carcinoma (OSCC). These samples were collected directly after surgical resection and were processed without formalin fixation. The dataset was split into approximately 50% tumor-positive and 50% tumor-free samples. For evaluation, these SRH images were compared against H&E-stained frozen sections, which served as the "gold standard" for intraoperative diagnosis.

Key Insights and Ideas

  • Label-Free and Stain-Free Imaging: SRH's ability to generate histopathology-like images without the need for chemical stains or tissue processing is a significant advancement, potentially streamlining intraoperative workflows.
  • Mimicking H&E Appearance: The "pseudo H&E" images generated by SRH are visually similar to conventional H&E slides, facilitating interpretation by pathologists familiar with traditional histology.
  • Potential for Real-Time Diagnosis: By eliminating sectioning and staining, SRH offers the possibility of much faster turnaround times for tissue analysis, which is critical in surgical settings.
  • Beyond Binary Classification: While SRH excels at distinguishing cancerous from non-cancerous tissue, its utility extends to sub-classifying various non-neoplastic tissues, providing richer diagnostic information.
  • Digital Data for AI: The inherently digital nature of SRH images makes them ideally suited for integration with artificial intelligence (AI) algorithms for automated analysis and potentially improved diagnostic accuracy.

Key Results

  • OSCC Detection: SRH demonstrated high accuracy in differentiating neoplastic (OSCC) from non-neoplastic tissue, with a Cohen's kappa of 0.880. The accuracy metrics were impressive: sensitivity of 100%, specificity of 90.91%, positive predictive value (PPV) of 90.00%, and negative predictive value (NPV) of 100%. The area under the receiver operating characteristic curve (AUC) was 0.954.
  • Non-Neoplastic Tissue Classification: SRH showed high agreement and accuracy for classifying normal mucosa, muscle tissue, and salivary glands. However, its performance varied for other non-neoplastic tissues like connective tissue and adipose tissue, where histological architecture differences were more pronounced.
  • Agreement with H&E: Overall, high agreement was observed between SRH and H&E-stained frozen sections for most tissue types, with kappa values ranging from 0.3246 (inflammatory cells) to 1.000 (lymph nodes).
  • Image Quality: Reduced image quality was observed in some SRH images, particularly in larger samples, which could impact diagnostic accuracy. However, these effects were not statistically significant in this study.

Limitations and Future Work

The authors acknowledge several limitations:

  • Resolution: The resolution of SRH is currently considered relatively low compared to conventional microscopy, which can make it difficult to distinguish subtle differences, particularly for precursor lesions or early-stage neoplastic changes.
  • Scan Area Limitation: The current maximum scan area of the device (approximately 25 mm²) can be a limitation for very large specimens, potentially leading to artifacts due to uneven thickness and autofocus issues.
  • Depth of Analysis: SRH scanned at a depth of 10 µm, while H&E frozen sections were prepared at a depth of approximately 100 µm. While the authors believe this did not introduce major bias due to the size of the evaluated structures, it means the exact same tissue layer wasn't analyzed.
  • Observer Bias: The study design, where all patients had a confirmed OSCC diagnosis, might have influenced the observers' interpretation, potentially contributing to the high sensitivity and specificity observed.

Future work should focus on:

  • Developing Tissue-Specific Protocols: Establishing optimized processing protocols for SRH, particularly for oral tissues, is crucial for its successful clinical implementation.
  • Increasing Pathologist Familiarity: Providing pathologists with more exposure and training in SRH interpretation is essential.
  • Hybrid Approaches: Combining conventional histological methods with SRH could facilitate its adoption.
  • AI Integration: Leveraging the digital nature of SRH images for AI-based analysis holds significant promise for further improving diagnostic capabilities.
  • Addressing Image Quality: Future iterations of the technology should aim to improve image quality, potentially through updated software for multi-scanning and advanced focusing capabilities.

Summary updated

Abstract

This study demonstrates the potential of SRH for intraoperative imaging of fresh, unprocessed tissue specimens from OSCC patients without the need for sectioning or staining.
BibTeX
@inproceedings{steybe2023stimulated,
  author = {David Steybe and Philipp Poxleitner and Marc Christian Metzger and René Rothweiler and Jürgen Beck and Jakob Straehle and Kirstin Vach and Andreas Weber and Kathrin Enderle-Ammour and Martin Werner and Rainer Schmelzeisen and Peter Bronsert},
  title = {{Stimulated Raman histology for histological evaluation of oral squamous cell carcinoma}},
  booktitle = {Clinical},
  year = {2023},
  doi = {10.1007/s00784-023-05098-9},
}
Figure