Application of Artificial Intelligence in Early Gastric Cancer Diagnosis

Abstract

<b><i>Background:</i></b> With the development of new technologies such as magnifying endoscopy with narrow band imaging, endoscopists achieved better accuracy for diagnosis of gastric cancer (GC) in various aspects. However, to master such skill takes substantial effort and could be difficult for inexperienced doctors. Therefore, a novel diagnostic method based on artificial intelligence (AI) was developed and its effectiveness was confirmed in many studies. AI system using convolutional neural network has showed marvelous results in the ongoing trials of computer-aided detection of colorectal polyps. <b><i>Summary:</i></b> With AI’s efficient computational power and learning capacities, endoscopists could improve their diagnostic accuracy and avoid the overlooking or over-diagnosis of gastric neoplasm. Several systems have been reported to achieved decent accuracy. Thus, AI-assisted endoscopy showed great potential on more accurate and sensitive ways for early detection, differentiation, and invasion depth prediction of gastric lesions. However, the feasibility, effectiveness, and safety in daily practice remain to be tested. <b><i>Key messages:</i></b> This review summarizes the current status of different AI applications in early GC diagnosis. More randomized controlled trails will be needed before AI could be widely put into clinical practice.

A Framework to Predict Gastric Cancer Based on Tongue Features and Deep Learning

“…However, the detection results are affected by the work experience and subjective perception of physicians. To overcome these limitations, researchers [ 25 , 26 , 27 ] built gastric cancer identification models to detect gastric cancer in endoscopic images by using artificial intelligence techniques. The model was trained based on a large number of endoscopic images of gastric cancer annotated by professional physicians.…”

Endoscopic methods for the detection and treatment of gastric cancer

“…CNN are tested on nonlabeled external datasets to determine whether the model can correctly identify previously unseen neoplasms. This process has been shown to significantly reduce error rates [22]. Furthermore, image enhancing methods, such as 3D image reconstruction, capsule chromo-endomicroscopy, and improvements in the image resolution using de-noising and de-blurring processes have improved the quality of images, further increasing the diagnostic accuracy and future utility of capsule endoscopy [23].…”

“…Furthermore, image enhancing methods, such as 3D image reconstruction, capsule chromo-endomicroscopy, and improvements in the image resolution using de-noising and de-blurring processes have improved the quality of images, further increasing the diagnostic accuracy and future utility of capsule endoscopy [23]. The greatest concern with the use of artificial intelligence are false negatives; this would be offset by imputation of enough high-quality images to its image bank and validation via use of multicenter randomized control trials, the latter of which is not yet available [22].…”

Accurate prediction of histological grading of intraductal papillary mucinous neoplasia using deep learning

“…Machine learning is a rapidly evolving scientific field with many applications in gastrointestinal endoscopy, such as colon poly detection [9][10][11], evaluation of Barrett's esophagus [12][13][14], and detection of early gastric cancer [15][16][17] or autoimmune pancreatitis [18]. However, literature using machine learning from EUS images to help classify pancreatic cysts is still sparse.…”