A Robust Spectro-Histopathological Diagnostic Method for Improved Gastric Cancer Detection

Area of Science:

• Biomedical Optics
• Cancer Diagnostics
• Artificial Intelligence in Medicine

Background:

• Gastric cancer is a leading cause of mortality, requiring fast intraoperative diagnostic tools.
• Autofluorescence spectroscopy offers a noninvasive method for tissue analysis but faces challenges in standardization, spectral resolution, and sample integrity.
• Existing methods lack the speed and accuracy needed for real-time intraoperative assessment.

Purpose of the Study:

• To develop and validate an integrated diagnostic method for rapid, accurate intraoperative gastric cancer detection.
• To overcome key challenges in autofluorescence spectroscopy for clinical application.
• To enhance AI-driven cancer diagnostic support systems.

Main Methods:

• Developed a spectral transfer model to standardize measurements between different spectrometers (R² = 0.977).
• Implemented a custom cooling chamber to maintain biological sample integrity during imaging.
• Utilized a modified U-Net 3+ model for precise tissue classification and segmentation, achieving 88.1% accuracy.

Main Results:

• The spectral transfer model successfully converted low-sensitivity spectra to high-quality, comparable spectra.
• The cooling chamber effectively preserved sample integrity during autofluorescence analysis.
• The U-Net 3+ model demonstrated high accuracy in tissue classification, validated against Raman microspectroscopy and H&E staining.

Conclusions:

• The integrated methodology enables rapid and reliable tissue analysis for gastric cancer.
• This approach addresses critical limitations of current autofluorescence spectroscopy techniques.
• The system shows significant potential for AI-enabled intraoperative cancer diagnostic support.