Comparative study on artificial intelligence systems for detecting early esophageal squamous cell carcinoma between narrow-band and white-light imaging

Figure 1 Flowchart of the procedures. CAD: Computer-assisted detection; NBI: Narrow-band imaging; WLI: White -light imaging.

Figure 2 Receiver operating characteristic curve for the test dataset. The area under the curve (AUC) was above 97%.

Figure 3 A comparison between the computer-assisted detection-narrow-band imaging and computer-assisted detection-white-light imaging systems in detecting early esophageal squamous cell carcinoma. ^aP < 0.05; ^bP < 0.01; ^cP < 0.001. CAD: Computer-assisted detection; NBI: Narrow-band imaging; WLI: White-light imaging.

Figure 4 Examples of computer-assisted detection system-diagnosed images. A and B: under white-light imaging (WLI) and narrow-band imaging (NBI), computer-assisted detection (CAD)-WLI and CAD-NBI recognized the esophageal cancer lesion (blue square); C and D: CAD-WLI mistakenly identified the normal mucosa as a lesion (blue square) in WLI, while CAD-NBI corrected it in NBI.

Figure 5 Improved accuracy of diagnosis with the assistance of the two computer-assisted detection systems according to the groups. A: Improvement of endoscopists’ accuracy in the second phase with the assistance of computer-assisted detection (CAD)-white-light imaging; B: Improvement of endoscopists’ accuracy in the third phase with the assistance of CAD-narrow-band imaging. WLI: White-light imaging.

Figure 6 Average diagnostic performance of the three groups of endoscopists in four phases. A: Accuracy; B: Sensitivity; C: Specificity. ^aP < 0.05; ^cP < 0.001.