Efficacy of Digestive Endoscope Based on Artificial Intelligence System in Diagnosing Early Esophageal Carcinoma

Abstract

Objective. To explore the efficacy of digestive endoscopy (DEN) based on artificial intelligence (AI) system in diagnosing early esophageal carcinoma. Methods. The clinical data of 300 patients with suspected esophageal carcinoma treated in our hospital from January 2018 to January 2020 were retrospectively analyzed; among them, 198 were diagnosed with esophageal carcinoma after pathological examination, and 102 had benign esophageal lesion. An AI system based on convolutional neural network (CNN) was adopted to assess the DEN images of patients with early esophageal carcinoma. A total of 200 patients (148 with early esophageal carcinoma and 52 with benign esophageal lesion) were selected as the learning group for the Inception V3 image classification system to learn; and the rest 100 patients (50 with early esophageal carcinoma and 50 with benign esophageal lesion) were included in the diagnosis group for the Inception V3 system to assist the narrow-band imaging (NBI) with diagnosis. The diagnosis results from Inception V3-assisted NBI were compared with those from imaging physicians, and the diagnostic efficacy diagram was drawn. Results. The diagnosis rate of AI-NBI was significantly faster than that of physician diagnosis ( 0.02 ± 0.01 vs. 5.65 ± 0.32 s (mean rate of two physicians), P < 0.001 ); between AI-NBI diagnosis and physician diagnosis, no statistical differences in sensitivity (90.0% vs. 92.0%), specificity (92.0% vs. 94.0%), and accuracy (91.0% vs. 93.0%) were observed ( P > 0.05 ); and according to the ROC curves, AUC (95% CI) of AI-NBI diagnosis = 0.910 (0.845-0.975), and AUC (95% CI) of physician diagnosis = 0.930 (0.872-0.988). Conclusion. CNN-based AI system can assist NBI in screening early esophageal carcinoma, which has a good application prospect in the clinical diagnosis of early esophageal carcinoma.

Efficiency of endoscopic artificial intelligence in the diagnosis of early esophageal cancer

“…Finally, 19 papers relating to AI‐assisted diagnosis of early EC and its invasion depth were included in the study. 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 By analyzing endoscopic WLI/NBI, the sensitivity and specificity of AI‐assisted systems, novices, and experts in the diagnosis of early EC and its infiltration depth were compared in these studies, respectively. The literature search process is shown in Figure 1 , and the basic information of each study was extracted (Table 1 ).…”

“…Subsequently, after reading through the full text of 178 publications, 159 were further excluded (reviews [49], detailed data were not provided [69], the number of people included in the study was less than 10 [28], and meta‐analysis [13]). Finally, 19 papers relating to AI‐assisted diagnosis of early EC and its invasion depth were included in the study 6–24 . By analyzing endoscopic WLI/NBI, the sensitivity and specificity of AI‐assisted systems, novices, and experts in the diagnosis of early EC and its infiltration depth were compared in these studies, respectively.…”

“…However, data from some studies have shown a certain number of false positives in AI diagnosis, thus resulting in over‐treatment. False positives were mainly found in endoscopic images regarding normal anatomy of the esophagogastric junction and cone compression, as well as benign lesions such as Barrett's esophagus, inflammation, and mucosal atrophy 24 . Therefore, it is important to train the AI system to learn more about images of tiny lesions to better reduce the false positive rate.…”

Machine Learning Applications for Early Detection of Esophageal Cancer: A Systematic Review

“…In the eld of early detection of EC, different imaging modalities such as gastroscopy, WLI, and narrow-band imaging (NBI) have been used in various studies (19)(20)(21). A review of the literature showed that WLI images were used in 35% of studies (11,17,(22)(23)(24)(25)(26)(27)(28)(29)(30)(31)(32)(33), followed by a combination of WLI and NBI images in 10% (18, 23,34,35), CT images in 13% (36-39), NBI images in 3% (40), images of other modalities in 13% (41)(42)(43)(44), and the type of imaging was not mentioned in 26% of studies (Fig. 4) (32,42,(45)(46)(47)(48)(49)(50).…”

“…Of the 15 studies reviewed, 11 employed various CNN algorithms for both segmentation and classi cation. The remaining 4 studies utilized other ML algorithms, including 2 studies of the MTL algorithm (26, 32), one study of the Neuro-T algorithm(25), and one study that used Google Net and TensorFlow algorithms(41). Among the studies, the highest accuracy was achieved by Wu et al(24) with a value of 96.28%, utilizing the Faster-RCNN and DSN algorithms for segmentation and classi cation.…”