A comparative study of shape features for polyp detection in wireless capsule endoscopy images

Area of Science:

• Medical Imaging
• Computer Vision
• Gastroenterology

Background:

• Wireless capsule endoscopy (WCE) enables direct visualization of the small bowel.
• The high volume of WCE images presents a significant challenge for manual polyp detection by clinicians.
• Automated polyp detection is crucial for efficient and accurate WCE analysis.

Purpose of the Study:

• To compare the efficacy of different shape features for detecting intestinal polyps in WCE images.
• To evaluate MPEG-7 region-based shape descriptors and Zernike moments for polyp recognition.
• To determine the optimal shape feature set for automated polyp detection in WCE.

Main Methods:

• Comparative analysis of MPEG-7 region-based shape descriptors and Zernike moments.
• Utilized a multi-layer perceptron neural network as the classification model.
• Trained and tested the model on a dataset of WCE images containing intestinal polyps.

Main Results:

• Both Zernike moments and MPEG-7 region-based shape descriptors show promise for recognizing intestinal polyp regions.
• Zernike moments demonstrated superior performance compared to MPEG-7 based features in polyp detection.
• The study validates the potential of shape-based features for automated WCE polyp analysis.

Conclusions:

• Shape features, particularly Zernike moments, are effective for automated intestinal polyp detection in WCE.
• Zernike moments offer a promising approach for improving the efficiency and accuracy of WCE data interpretation.
• Further research can leverage these findings to develop advanced WCE analysis tools.