In this prospective paired diagnostic study and single-center, randomized controlled trial, patients with early esophageal squamous neoplasia or high-grade intraepithelial neoplasia meeting the inclusion and exclusion criteria will be enrolled in a paired diagnostic cohort (60 patients) and subsequently randomly assigned (1:1) to receive endoscopic submucosal dissection (ESD) with AI-based Dissection Trajectory Prediction System (ADTPS) guidance or conventional ESD (without AI). Clinical data and operator workload scores (NASA-TLX) are collected during the procedure, and postoperative follow-up assessments are performed at days 1, 3, 7, and 14. The study aims to analyze the impact of ADTPS on the mean single-dissection time and operator workload in patients undergoing ESD by comparing the efficacy differences between the experimental and control groups. Additionally, the study investigates the effects of ADTPS on other postoperative complications including R0 resection rate, muscularis propria injury, intraoperative bleeding, perforation (acute and delayed), and total procedure time; conducts a comparative analysis of the safety and efficiency of AI-assisted versus conventional ESD; and develops effective clinical strategies for optimizing dissection trajectory and reducing complications in endoscopic submucosal dissection.