A deep convolutional neural network for bleeding detection in Wireless Capsule Endoscopy images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Wireless Capsule Endoscopy (WCE) is crucial for non-invasive small bowel examination.
• Computer-aided diagnosis (CAD) for gastrointestinal (GI) bleeding in WCE aims to reduce physician workload.
• Current handcrafted feature methods lack sufficient accuracy for reliable bleeding detection.

Purpose of the Study:

• To develop an automatic bleeding detection strategy for WCE images.
• To enhance the accuracy of gastrointestinal bleeding detection using deep learning.
• To evaluate the performance of the proposed method on a large WCE image dataset.

Main Methods:

• Implementation of a deep convolutional neural network (CNN) for bleeding detection.
• Training and evaluation on an expanded dataset of 10,000 WCE images.
• Comparison with existing state-of-the-art WCE bleeding detection approaches.

Main Results:

• The proposed deep learning method achieved a significant increase in detection accuracy.
• An improvement of approximately 2 percentage points in the F1 score was observed.
• The method demonstrated superior performance compared to existing WCE bleeding detection techniques.

Conclusions:

• The deep convolutional neural network strategy offers a more effective approach to WCE bleeding detection.
• This advanced method shows potential for improving diagnostic efficiency and accuracy in gastroenterology.
• The F1 score reached up to 0.9955, indicating high performance in detecting GI bleeding.