Low Complexity CNN Structure for Automatic Bleeding Zone Detection in Wireless Capsule Endoscopy Imaging.

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Wireless capsule endoscopy (WCE) aids digestive system screening.
• Automatic WCE analysis using CNNs is prevalent but computationally complex.
• Existing methods often overlook CNN structural and computational efficiency.

Purpose of the Study:

• To propose a simplified, low-complexity CNN structure for WCE image analysis.
• To address the computational burden of current automatic WCE screening methods.
• To enhance the efficiency and accuracy of bleeding zone detection in WCE.

Main Methods:

• Developed a simplified CNN inspired by the Fully Convolutional Network (FCN) paradigm.
• Input: single image patch; Output: segmented patch of the same size.
• Employed non-overlapping patch processing to circumvent redundant computations and utilize moderate feature maps.

Main Results:

• The proposed network achieved higher accuracy and AUC compared to previous structures.
• Demonstrated a significant reduction in computational operations.
• Successfully evaluated on the publicly available KID dataset for WCE image analysis.

Conclusions:

• The simplified CNN offers an efficient and accurate solution for WCE bleeding zone detection.
• This approach reduces computational complexity without compromising performance.
• Paves the way for more practical and widespread adoption of automated WCE analysis.