br In this study the results of tests and were
In this study, the results of tests 1 and 3 were compared; then, the significant improvement in the diagnostic accu-racy of M-NBI after the e-learning training was confirmed. In this secondary analysis, analysis of learning effects was performed by dividing the lesion into fine features. In addi-tion, no reports evaluated the AUC/ROC of detailed find-ings obtained by magnifying endoscopy using NRI and IDI analysis. In the result of the AUC/ROC analysis accord-ing to lesion shape, a significant learning effect was demon-strated in depressed lesions (AUC/ROC, .90 in test 3). In the comparison of the AUC of DL, depressed lesion (AUC, .75; test 3) had higher AUC than that of flat and elevated lesions (AUC, .55; test 3), and DL was highly use-ful in depressed lesions. Based on these findings, evalu-ating DL in a depressed lesion is useful for differentiating between cancerous and noncancerous lesions. In addition, EGC often exhibits a depressed macroscopic type. Hence, M-NBI diagnosis is considered to be highly useful for many EGCs. Meanwhile, the AUC/ROC of elevated and flat le-sions was relatively low (.69 and .63, respectively). Further-more, no statistically significant increase in AUC/ROC was observed when comparing before and after e-learning training. This suggests that M-NBI observation evaluates
www.giejournal.org Volume -, No. - : 2019 GASTROINTESTINAL ENDOSCOPY 5
Learning effect after e-learning training in endoscopic diagnosis Ikehara et al
Figure 3. Changes in accuracy between tests 1 and 3 in 365 participants.
TABLE 3. AUC/ROC comparison between tests 1 and 3
AUC/ROC, Area under the receiver operating characteristic curve; CI, confidence interval; cNRI, continuous net reclassification improvement; IDI, integrated discrimination improvement.
the lesion surface layer; thus, it o-Phenanthroline tends to be lower in elevated and flat lesions, which are thicker than the depressed type.
In the study by size, AUC/ROC tended to be high in small lesions <10 mm, and the learning effect of the e-learning training showed the highest result (test 1, .79;
test 3, .93; cNRI, 1.46; IDI, .19). In the examination before and after e-learning training in lesions 10 mm, although a significant increase in cNRI was observed, there was no sig-nificant improvement in IDI. In addition, the AUC/ROC showed a trend lower by 10 mm. From these results, improvement of M-NBI diagnostic ability by e-learning
6 GASTROINTESTINAL ENDOSCOPY Volume -, No. - : 2019 www.giejournal.org
Ikehara et al
Learning effect after e-learning training in endoscopic diagnosis
TABLE 4. AUC/ROC of DL, MV, and MS comparison between tests 1 and 3
AUC/ROC, area under the receiver operating characteristic curve; DL, demarcation line; MV, microvascular pattern; MS, microsurface pattern; CI, confidence interval; cNRI, continuous net reclassification improvement; IDI, integrated discrimination improvement.
No. of lesions Fleiss’ kappa Interpretation
Fleiss’ kappa Interpretation
Inflammation 16 .20 Slight .26 Fair
Elevated lesion 10 .22 Fair .24 Fair
Flat lesion 6 .20 Slight .24 Fair
Depressed lesion 24 .30 Fair .38 Fair
*Others include adenoma and xanthoma.
training is more likely to be obtained in small-sized cancers of depressed type.
White opaque substance (WOS) is lipid droplets or fat droplets absorbed and stored in the outermost layer of the epithelium.15 In the presence of WOS, the projection light is strongly scattered; hence, the visibility of blood vessels under the epithelium is reduced. WOS indicates the phenomenon observed as a white border as a result of the above.16 WOS is observed in gastric mucosa and in gastric adenoma that acquired intestinal phenotype. WOS is reported to be useful in differentiating between adenoma and cancer.15 In this study, accuracy was 49.6% and 64.3% in those WOS-positive lesions (Supplementary Table 1, available online at www.giejournal.org). In contrast, accuracy before and after e-learning training in WOS-negative lesions was 64.0% and 69.0%, respectively (Supplementary Table 2, available online at www.giejournal.org). In contrast, accuracy was 49.6% and 64.3% in those WOS-positive lesions (Supplementary Table 1, available online at www.giejournal.org). Nevertheless, a significant learning effect was obtained in those 2 groups, and making accurate diagnosis in WOS-positive lesions appears to be difficult.
Gastric cancer develops from gastric mucosa affected by chronic gastritis caused by Helicobacter pylori. There-fore, when diagnosing gastric lesions, Slow component is important to distinguish between gastric cancer and gastritis. In the ex-amination of the diagnostic variation by kappa value, the kappa value had “slight agreement” in inflammatory le-sions before e-learning training (test 1), indicating a low concordance rate. Because gastric mucosa with chronic gastritis may show an irregular blood vessel image in M-NBI, diagnosis varies among evaluators.