Gastric cancer remains one of the leading causes of cancer-related deaths worldwide, necessitating a rapid intraoperative method to determine the presence or absence of malignancy. Autofluorescence spectroscopy has emerged as a promising noninvasive diagnostic tool, utilizing endogenous fluorophores to analyze tissue characteristics. However, three major challenges persist: standardizing instrument-to-instrument variance in spectroscopic measurements, resolving overlapping spectral signatures from different fluorophores, and maintaining the biological sample integrity during optical analysis. To address these challenges systematically, this work proposes an integrated diagnostic method with quantitative validation. We developed a spectral transfer model (R2 = 0.977) trained on corresponding spectral pairs from low- and high-sensitivity spectrometers, enabling the conversion of low-sensitivity measurements into high-quality spectra comparable to those from high-sensitivity instruments. Additionally, a custom-designed cooling chamber was implemented to preserve biological sample integrity during autofluorescence imaging. For precise tissue classification and segmentation, we employed a modified U-Net 3+ model, achieving an overall accuracy of 88.1%. The efficacy of our integrated methodology was rigorously validated through comparison with Raman microspectroscopy and standard hematoxylin and eosin (H&E) staining of tissue sections. This integrated methodology enables rapid tissue analysis and demonstrates significant potential for AI-enabled cancer diagnostic support systems.
Jang et al. (Wed,) studied this question.