Table 1.
Number of images collected for WLI/NBI data.
Fig 1.
Labeling data for regions of interest.
(a) white-light imaging (WLI). (b) narrow-band imaging (NBI).
Fig 2.
Architecture for Tumor detection in esophageal Endoscopy images.
(a) YOLOv5. (b) RetinaNet.
Table 2.
Performance evaluation metrics for detection models based on confidence thresholds from internal data.
Fig 3.
TP predictions from a trained model for tumor location detection.
(a–d) YOLOv5, (e–h) RetinaNet. (blue color: ground truth, red color: predicted result).
Fig 4.
Prediction results of a trained model for detecting the location of a tumor.
(a, b) False positive, FP. (c, d) False negative, FN. (blue color: ground truth, red color: predicted result).
Fig 5.
Precision–recall curves obtained using the detection model for internal data.
(a) YOLOv5. (b) RetinaNet.
Table 3.
Performance evaluation metrics for detection models based on confidence thresholds from external data.
Fig 6.
TP predictions from a trained model for tumor location detection.
(a–d) YOLOv5, (e–h) RetinaNet. (blue color: ground truth, red color: predicted result).
Fig 7.
Prediction results of a trained model for detecting the location of a tumor.
(a, b) False positive, FP. (c, d) False negative, FN. (blue color: ground truth, red color: predicted result).
Fig 8.
Precision–recall curves obtained using the detection model for external data.
(a) YOLOv5. (b) RetinaNet.