Gastric polyp detection module based on improved attentional feature fusion.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Gastric cancer is a significant health concern, with gastric polyps posing a high risk of malignancy.
• Early detection of gastric polyps is crucial for reducing gastric cancer incidence.
• Deep learning-based object detection is common in medical imaging, but struggles with low-contrast endoscopic images.

Purpose of the Study:

• To improve the detection performance of gastric polyps in endoscopic images.
• To address challenges posed by poor contrast and varying polyp sizes in gastroscopic images.

Main Methods:

• Proposed an attention module to enhance polyp-background contrast and focus on relevant features.
• Developed a novel attentional feature fusion module integrating attention with multi-scale feature fusion.
• Conducted experiments using a custom dataset of gastric polyps.

Main Results:

• The proposed attentional feature fusion module demonstrated superior performance compared to conventional feature fusion methods.
• The method effectively improved the detection of gastric polyps, reducing missed and misdetected instances.
• Enhanced contrast and feature highlighting led to more accurate polyp identification.

Conclusions:

• The developed attentional feature fusion module significantly enhances the accuracy of endoscopic gastric polyp detection.
• This approach offers a promising solution for improving early diagnosis and management of gastric conditions.
• The method shows potential for clinical application in reducing gastric cancer rates through timely polyp detection.