ARTIFICIAL INTELLIGENCE IN ENDOSCOPY FOR GASTRIC CANCER ASSESSMENT: MACHINE OVER MAN OR PERFECT HARMONY?

Aims Artificial intelligence (AI) is likely to execute roles currently performed by humans, being essential for endoscopists to focus on this novel technology for avoiding to miss and mischaracterize malignant change in the stomach. The aim of our study was to assess the advances of AI-based medicine and the involvement in rectifying current limitations of gastroscopy.

Methods A systematic literature search was carried out in three major databases which are as follows: PubMed, Scopus, and Embase up to November 2019. The analysis was performed using the population intervention comparison outcome (PICO) format: (P) patients undergoing endoscopy for the assessment of gastric cancer; artificial intelligence (I) over endoscopist (C), the outcome (O) being the diagnostic accuracy.

Results 16 papers were selected for the systematic review. Computer-aided diagnosis, convolutional neural network computer-aided detection platforms with or without magnifying endoscopy with narrow band imaging, magnifying endoscopy with blue-laser imaging, support vector machine-based analysis system have been described in a number of pilot studies showing excellent results. The systems were able to determine quantitatively the early gastric cancers with deep submucosal invasion, to minimize the overestimation of invasion depth, to achieve a high diagnostic accuracy in detecting gastric cancers with a sensitivity equivalent to that of expert endoscopists and surpassing the non-expert endoscopists. AI can detect features in medical images that the human eye cannot assess or even see (hyperspectral imaging). Overall, AI systems using deep learning algorithms achieved a remarkable progress in medical imaging especially in colonic diseases but their application in other parts of the gastrointestinal tract has been limited.

Conclusions AI is expected to provide on-site decision support and help endoscopists, regardless of their skill, deliver a more accurate diagnosis during real time endoscopy by automatically detecting and categorizing lesions. Hence, it is essential that endoscopists focus on this novel technology and act in perfect harmony.