The value of artificial intelligence-assisted cytology in the diagnosis of esophageal cancer and precancerous lesions
10.3760/cma.j.cn321463-20240606-00141
- VernacularTitle:人工智能辅助细胞学诊断食管癌及癌前病变的价值
- Author:
Min XU
1
;
Lin LIU
1
;
Zhiheng CHANG
1
;
Bofu TANG
1
;
Tong DANG
1
Author Information
1. 包头医学院第二附属医院消化内科 内蒙古消化病研究所，包头 014030
- Publication Type:Journal Article
- Keywords:
Esophageal neoplasms;
Artificial intelligence-assisted cytology;
Cytosponge;
Gastroscopy;
Diagnostic value
- From:
Chinese Journal of Digestive Endoscopy
2025;42(4):314-318
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To evaluate the diagnostic efficacy of artificial intelligence-assisted cytology in detecting esophageal cancer and precancerous lesions.Methods:Outpatients who were scheduled for endoscopic examination at the Second Affiliated Hospital of Baotou Medical College from June 21, 2021 to May 12, 2022 were selected. Artificial intelligence-assisted cytology and painless gastroscopy (add narrow-band imaging and biopsy pathology when need) (hereinafter referred to as gastroscopy) were performed. The detection rates of esophageal cancer and precancerous lesions by the two methods were compared. Taking gastroscopy as the standard, sensitivity, specificity, positive predictive value, negative predictive value of artificial intelligence-assisted cytology diagnosis were calculated to verify its diagnostic efficacy.Results:A total of 849 patients meeting the inclusion criteria were included. Artificial intelligence-assisted cytology detected 365 cases of esophageal lesions, with a detection rate of 42.99%. One hundred cases of esophageal cancer, precancerous lesions, and atypical squamous cell hyperplasia were detected, with a detection rate of 11.78%. Gastroscopy detected 281 cases of esophageal lesions, with a detection rate of 33.10%. Eighty cases of esophageal cancer and precancerous lesion were detected, with a detection rate of 9.42%. The sensitivity and specificity of artificial intelligence-assisted cytology diagnosis were 81.25% (65/80) and 95.45% (734/769), respectively. The positive predictive value was 65.00% (65/100), and the negative predictive value was 98.00% (734/749). The area under the receiver operating characteristic curve of artificial intelligence-assisted cytology in dectecting esophageal cancer and precancerous lesions was 0.785.Conclusion:Artificial intelligence-assisted cytology demonstrates relatively high sensitivity and specificity in diagnosing esophageal cancer and precancerous lesions, exhibiting good diagnostic performance for esophageal lesions.
