an AI-Driven Four-Tier Invasion Depth Classification for Early Esophageal Cancer

an AI-Driven Four-Tier Invasion Depth Classification for Early Esophageal Cancer

The goal of this retrospective observational study is to evaluate whether preoperative endoscopic imaging can accurately assess tumor invasion depth in patients with early esophageal neoplasia undergoing endoscopic submucosal dissection (ESD).

The main question it aims to answer is:"How accurately can preoperative multimodal endoscopic imaging predict histopathological invasion depth in early esophageal squamous cell carcinoma and high-grade intraepithelial neoplasia?" If there is a comparison group: Not applicable (no intervention or arm comparison was specified; analyses are based on diagnostic performance against postoperative pathology as the reference standard).

Participants will:

• Be retrospectively identified from two tertiary hospitals in China;
• Have pathologically confirmed early esophageal squamous cell carcinoma or high-grade intraepithelial neoplasia treated with ESD;
• Have complete postoperative pathology data including invasion depth, lesion size, location, lymphovascular invasion, and margin status;
• Have preoperative high-quality endoscopic images (white-light imaging, narrow-band imaging, iodine staining, and blue laser imaging);
• Undergo retrospective image-pathology correlation analysis to assess diagnostic performance of invasion depth assessment.

Inclusion Criteria:

• Pathologically confirmed EEC or high-grade intraepithelial neoplasia;
• Underwent ESD with complete postoperative pathological report;
• Preoperative high-quality WLI, NBl, iodine staining and BLI endoscopic images;
• complete clinical data.

Exclusion Criteria:

• Poor-quality endoscopic images;
• Previous esophageal surgery, neoadjuvant chemoradiotherapy, or other malignant tumors;
• Missing pathological invasion depth data;
• Synchronous distant metastasis.