Gastrointestinal bleeding detection in wireless capsule endoscopy images using handcrafted and CNN features

Area of Science:

• Medical Imaging
• Gastroenterology
• Artificial Intelligence

Background:

• Gastrointestinal (GI) bleeding detection is crucial for wireless capsule endoscopy (WCE).
• Accurate bleeding detection improves patient diagnosis and treatment planning.
• Existing methods may face challenges with limited training data or high computational costs.

Purpose of the Study:

• To develop a novel and efficient approach for GI bleeding detection in WCE.
• To combine handcrafted (HC) features with convolutional neural network (CNN) features.
• To reduce computational cost by using a smaller-scale CNN architecture.

Main Methods:

• A hybrid feature extraction method combining HC features and CNN features was developed.
• A smaller-scale CNN architecture was designed to optimize computational efficiency.
• The proposed method was evaluated for its performance, particularly with limited training data.

Main Results:

• The proposed WCE bleeding detection strategy demonstrated high capability with limited training data.
• Experimental results showed that the method achieved comparable or superior performance against state-of-the-art techniques.
• The reduced CNN architecture effectively lowered the computational cost without compromising accuracy.

Conclusions:

• The combined HC and CNN feature approach offers a promising solution for WCE GI bleeding detection.
• The developed method is robust and efficient, especially in scenarios with scarce training datasets.
• This strategy presents a valuable advancement for improving WCE diagnostic capabilities.