Lightweight deep learning model for gastrointestinal precancerous lesion screening with attention enhancement

1

The study developed a lightweight attention-enhanced MobileNetV3 model for classifying gastrointestinal precancerous lesions.
2

The model achieved an overall accuracy of 99.10% and 100% precision for polyp detection, significantly improving diagnostic performance.
3

Ablation experiments confirmed the effectiveness of the Spatial-Channel Attention module, enhancing core metrics without increasing computational load.
4

The model's lightweight design allows real-time deployment on endoscopic devices, addressing barriers to clinical adoption of AI.
5

GradCAM visualizations demonstrated the model's focus on clinically relevant areas, aligning AI reasoning with endoscopists' observation habits.

1