Deep learning for diagnosis of precancerous lesions in upper gastrointestinal endoscopy: A review

Figure 1  Infographic with icons and timeline for artificial intelligence, machine learning and deep learning.

Figure 2 Illustration of the diagnostic process of physician, machine learning and deep learning. A: Physician diagnostic process; B: Machine learning; C: Deep learning. Conv: Convolutional layer; FC: Fully connected layer; GIM: gastric intestinal metaplasia.

Figure 3 Data augmentation for a typical magnifying narrow band image for training a convolutional neural network model. This is performed by using a variety of image transformations and their combinations. A: Original image; B: Flip horizontal and random rotation; C: Flip vertical and magnification; D: Random rotation and shift; E: Flip horizontal, minification and shift; F: Flip vertical, rotation and shift.

Figure 4 Informative features (partially related to lesions areas) acquired by the convolutional neural networks, where warmer colors mean higher contributions to decision making. A: Original endoscopic images; B: Corresponding attention. BE: Barrett’s esophagus; GIM: Gastric intestinal metaplasia.