A colorectal polyp image segmentation method, system, electronic device and medium
By using a colorectal polyp segmentation neural network based on the EfficientNet-B0 and U-Net frameworks, combined with probability distribution images and class activation heatmaps, the contradiction between high accuracy and low latency in existing technologies is resolved, enabling efficient and accurate polyp segmentation on clinical edge devices.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- INST OF MEDICAL INFORMATION CHINESE ACAD OF MEDICAL SCI
- Filing Date
- 2026-04-10
- Publication Date
- 2026-07-17
AI Technical Summary
Existing colorectal polyp segmentation algorithms struggle to simultaneously meet the requirements of high accuracy, low computational complexity, and low latency in real-time clinical deployments. Mainstream models incur high computational overhead and have long inference times, while lightweight models lack sufficient accuracy when handling small polyps or ambiguous boundaries, failing to reach clinically reliable levels.
A polyp segmentation neural network built using the EfficientNet-B0 encoder and the U-Net framework achieves pixel-level segmentation prediction and interpretable feature extraction through a single forward propagation, generating polyp probability distribution images and class activation heatmaps. This avoids additional gradient backpropagation or secondary network calls. Combined with binarized masks and heatmap outputs, it provides verifiable visual interpretations.
With low computational complexity and low latency, high colorectal polyp segmentation accuracy is achieved, meeting the real-time requirements of clinical edge devices and providing interpretable visual aids for diagnosis, thus improving the effective accuracy of the lightweight model.
Smart Images
Figure CN122415642A_ABST