An intelligent diagnostic algorithm for Raman spectroscopy of gastrointestinal cancer based on component modeling.

Early diagnosis of gastrointestinal (GI) cancer is crucial for patient prognosis, yet conventional methods suffer from invasiveness and insufficient molecular sensitivity. While Raman spectroscopy offers non-invasive molecular fingerprinting, spectral overlap in complex biological samples poses a challenge. To address this, this study introduces a diagnostic framework that synergizes Raman spectroscopy with a convolutional neural network (CNN) to quantitatively resolve spectral components for improved GI cancer detection. Based on Raman spectra from 829 GI tissues and five pure components (DNA, triolein, histone, collagen, and actin), an improved CNN regression model was trained on 100 000 simulated spectra. The model accurately quantified the relative proportions of these five biochemicals (R²: 0.91-0.98), revealing significantly higher coefficients for DNA, collagen, and actin, and lower coefficients for triolein and histone in malignant tissues (P < 0.01). Utilizing these quantitative molecular features, a subsequent LightGBM classification model achieved an accuracy of 97.2%, a sensitivity of 90%, a specificity of 98.1%, and an AUC of 0.973 on an independent test set. This work therefore demonstrates a powerful approach for discriminating benign and malignant GI tissues by quantitatively modeling key molecular alterations. The high classification accuracy validates the clinical translational potential of this non-invasive method for GI cancer screening and offers a generalizable strategy for other complex biological analyses and diagnostics.