AutoGERD: a multimodal deep-learning model integrating endoscopic images and esophageal physiological measurements for anti-reflux surgery assessment

De-identified datasets, redacted distal esophageal endoscopic images, and core Python code supporting the study: "AutoGERD: a multimodal deep-learning model integrating endoscopic images and esophageal physiological measurements for anti-reflux surgery assessment". Contents data/tables/: Structured clinical and esophageal physiological measurements (32 variables from HREM and 24-h MII-pH monitoring), alongside data_dictionary.csv and model_feature_order.csv: internal_functional_707.csv (n = 707) and internal_multimodal_662.csv (n = 662) from the primary center . external_patients_45.csv (n = 45) and external_multimodal_40.csv (n = 40) from the external validation center. data/images/: Redacted distal esophageal endoscopic images for the internal (internal/, n = 662) and external (external/, n = 40) multimodal cohorts. core_code/: Source code and dependencies (requirements.txt, README.md) for data preprocessing, feature selection, model training and evaluation (AUTOGERD-FM and AUTOGERD-MM), DeLong comparisons, and SHAP explainability analyses. (Note: Model weights and cached training image augmentations are not included.) Ethics Statement All clinical tables and endoscopic images have been de-identified and redacted to remove personal identifiers. Ethical approval was granted by the Ethics Committee of Yan’an Hospital Affiliated to Kunming Medical University (No. 2023-082-01), with written informed consent waived due to the retrospective study design.