ABSTRACT

Objective: Gastric cancer polyps is a common digestive system disease, which is closely related to gastric cancer. This study focuses on AI (Artificial Intelligence)-assisted gastroscopy for cancer polyp detection, including methodology exploration, brand-new system construction and real-time clinical trials, in order to realize the independent development and validity verification of AI-based cancer polyp collaborative diagnosis system and provide an efficient diagnostic means for gastroscopy for polyp detection. Methods: By studying the risk factors and experimental samples, the experimental samples were trained with BPNN (BP neural network) in the artificial neural network, and the conclusion was drawn. GA (genetic algorithm) is used to optimize BPNN. BPNN discriminant function is used as discriminant function for embedding. Results: The best threshold of AI model for cancer polyp detection was 0.79, and the AUC under the curve was 0.882. Combining AI white light mode with NBI(narrow band imaging) mode, we can find that the sensitivity of AI parallel analysis is the highest (88.8%) and the specificity of AI series analysis is the highest (95.238%). Conclusion: The real-time application of AI-based cancer polyp collaborative diagnosis system can improve the detection results of gastric cancer polyps to some extent, and it is safe and effective to use AI-assisted detection system to detect cancer polyps in real time.