Objective To construct a diagnostic model for spleen-stomach damp-heat syndrome in patients with chronic gastritis (CG) based on hyperspectral tongue imaging data, and to investigate the diagnostic value of hyperspectral tongue imaging in identifying spleen-stomach damp-heat syndrome in CG patients.

Methods Using the prototype V3.0 image-spectral data collection device, intelligent tongue surface observations were conducted on CG patients at The Mingyi Hall of The Second Affiliated Hospital of Anhui University of Chinese Medicine between September 4, 2024 and October 29, 2025. Patients were categorized into spleen-stomach damp-heat syndrome and non-spleen-stomach damp-heat syndrome groups, and randomly assigned to a training set and an internal validation set at a ratio of 7 : 3. Following preprocessing and segmentation of the tongue images, wavelength-specific spectral features were extracted. The least absolute shrinkage and selection operator (LASSO) algorithm was applied for dimensionality reduction and feature selection of hyperspectral tongue features, and a diagnostic model was constructed based on the selected features. Pearson correlation analysis was performed to assess the correlation between the diagnostic model and clinical data, including tongue and pulse manifestations, symptoms, gastroscopic findings, and lifestyle habits. The diagnostic performance of the model was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC), sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV). Five-fold cross-validation was performed on the internal validation set to calculate the mean AUC. Bootstrap calibration curves and decision curve analysis (DCA) based on 500 resamples were performed to evaluate model calibration and clinical net benefit.

Results A total of 383 CG patients were enrolled and classified into the spleen-stomach damp-heat syndrome group (n = 219) and the non-spleen-stomach damp-heat syndrome group (n = 164). The LASSO algorithm identified nine hyperspectral tongue features for inclusion in the diagnostic model. The diagnostic model was significantly correlated with 13 clinical variables (P < 0.05). It was positively correlated with yellow greasy coating, and negatively correlated with thin white coating, yellow coating, and scanty or absent coating (P < 0.001). It was positively correlated with slippery and rapid pulse, and negatively correlated with thready and rapid pulse and wiry pulse (P < 0.001). Diagnostic model was positively correlated with male sex, alcohol consumption, smoking, chronic atrophic gastritis (CAG) diagnosed by gastroscopy, heaviness of the body, and sticky stools (P < 0.05, P < 0.01, or P < 0.001). Model validation yielded an AUC of 0.777, an accuracy of 0.7043, a sensitivity of 0.7419, a specificity of 0.6604, a PPV of 0.7188, and an NPV of 0.6863 in the internal validation set. Five-fold cross-validation produced a mean AUC of 0.764. The Hosmer-Lemeshow goodness-of-fit test indicated that χ² = 5.4238 and P = 0.7115. Bootstrap DCA based on 500 resamples demonstrated, within the high-risk threshold range of 0.16 – 0.80, the model provided greater net benefit than either full therapy or no treatment.

Conclusion This study developed a diagnostic model for spleen-stomach damp-heat syndrome among CG patients based on tongue hyperspectral data which showed promising predictive performance, and the model was correlated with the traditional Chinese medicine (TCM) characteristics of spleen-stomach damp-heat syndrome, providing new evidence for the objective diagnosis of this syndrome.