Real-time colonoscopic detection and precise segmentation of colorectal polyps via PESNet.

Introduction: Precise and timely visual assistance is critical for detecting and completely removing colorectal cancer precursor polyps, a key step in preventing interval cancer and reducing patient morbidity. Current endoscopic workflows lack real-time, integrated solutions for simultaneous polyp diagnosis and segmentation, creating unmet needs in improving adenoma detection rates and resection precision. Methods: We propose PESNet, a real-time assistance framework for standard endoscopy workstations. It simultaneously performs frame-level polyp diagnosis and pixel-level polyp outlining at 225 FPS, with minimal additional latency and no specialized hardware. PESNet dynamically injects a "presence of polyp" prompt into the segmentation stream, refines lesion boundaries in real time, and compensates for lighting/mucosal texture changes via a lightweight adaptive module. Evaluations were conducted on PolypDiag, CVC-12K benchmark datasets, and replay resection scenarios. Latency was measured using TensorRT FP16 on an RTX 6000 Ada GPU. Results: On PolypDiag and CVC-12K, PESNet improved diagnostic F1 from 95.0% to 97.2% and segmentation Dice from 85.4% to 89.1%. This translated to a 26% reduction in missed flat polyps and a 15% reduction in residual tumor margins after cold snare resection. End-to-end latency (1080p) was 12.6 ± 0.3 ms per frame, with segmentation (4.4 ms), prompt fusion (0.6 ms), and prototype lookup (< 0.2 ms) all satisfying a 40 ms clinical budget with > 3× headroom. Discussion: These clinically significant improvements demonstrate PESNet's potential to enhance adenoma detection rates, support cleaner resection margins, and ultimately reduce colorectal cancer incidence during routine endoscopic examinations. Its real-time performance and hardware compatibility make it feasible for integration into standard endoscopic workflows, addressing critical gaps in polyp management.