Aims We have developed the artificial intelligence (AI) diagnostic system for the detection
of superficial esophageal squamous cell carcinoma (ESCC) applied deep learning system.
The aim of this study was to reveal the effect of AI diagnostic system on the improvement
of ESCC detection in clinical trial.
Methods This study was a prospective, single-center, exploratory randomized controlled trial.
From October 2020 to November 2021, 325 patients undergoing screening or surveillance
esophagogastroduodenoscopy with primary head and neck cancer and/or ESCC were enrolled.
All participants were randomly assigned to either AI group (n=155) or control group
(n=170). In both groups, expert and non-expert endoscopists screened the esophagus
with WLI followed by NBI with or without AI assistance. Then, after iodine chromoendoscopy
of the esophagus, we took biopsies from suspicious ESCCs detected by any of the modalities.
The primary endpoint was the additional effect on the detection rate of ESCC with
AI diagnostic system in non-experts.
Results Among 325 patients, 41 lesions of ESCC (21 in AI group, 20 in control group) were
diagnosed. Non-experts could detect 44% of ESCCs (7/16) in AI group, and 45% (5/11)
in control group. Experts could detect 87% of ESCCs (4/5) in AI group, and 57% (5/9)
in control group. No significant differences were observed in terms of detection sensitivity
in non-experts between the two groups (p=0.94), as well as in all endoscopists (53%
vs 50%, p=0.84).
Conclusions AI diagnostic system could not improve the detection rate of ESCC in this clinical
trial.
