AGCNet: A Precise Adaptive Global Context Network for Real-Time Colonoscopy

Colonic endoscopy is the gold standard for detecting rectal polyps and cancer. In which are a major predisposing factor colorectal cancer, precise diagnosis of within highly dependent on physician professional level. With development deep learning, some semantic segmentation methods have recently been applied to polyp detection, but there problems with insufficient accuracy speed. To this end, we propose precision adaptive global context network (AGCNet) based real-time colon endoscopy. Firstly, in order adapt problem large-scale variation polyps, designed multi-scale fusion module (MSFM), enhances representation capability by varieties filters collect contextual information at different scales, thus adapting large size, especially smaller polyps. addition, modelling long-range dependence simply using complex spatial pixels tends introduce more background noise increase computational effort. context-aware pyramid aggregation (CPAM) was designed, internally includes novel dual attention mechanism whereby CPAM aggregates feature across regions boost network’s ability utilize model dependency through further reinforce features important efficiently suppress non-important regions. Additionally, performs multi-level pooling input extract from image uses an selectively highlight informative that most relevant task. The fuses pooled map produce enhanced representations capture both local information. Thereby achieving taking into account. Our proposed AGCNet performed extensive experimental studies datasets Kvasir-SEG, CVC-ClinicDB, CVC-ColonDB ETIS-LaribPolypDB. Specifically, achieved IoU 87.40% Dice score 92.63% Kvasir dataset, accurate results faster than many current state-of-the-art models.