Color-Transfer-Enhanced Data Construction and Validation for Deep Learning-Based Upper Gastrointestinal Landmark...

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Wireless capsule endoscopy (WCE) is increasingly adopted but primarily used for small intestine imaging.
• Upper gastrointestinal tract examinations using WCE are anticipated to grow with technological advancements.
• Limited research and datasets exist for landmark identification in upper GI WCE.

Purpose of the Study:

• To develop a novel approach for anatomical landmark identification in the upper gastrointestinal tract using WCE.
• To create a simulated WCE dataset for upper GI tract examinations.
• To evaluate the accuracy of deep learning models for landmark classification in this context.

Main Methods:

• A simulated WCE dataset for the upper GI tract was created using color transfer techniques.
• Similarity between simulated and authentic WCE images was verified using Euclidean distance measurements.
• The DenseNet169 deep learning model was employed for anatomical landmark classification after image preprocessing and similarity evaluation.

Main Results:

• The developed approach achieved over 90% accuracy in anatomical landmark classification in the upper GI tract using the simulated dataset.
• Applying sharpen and detail filters improved classification accuracy from 91.32% to 94.06%.

Conclusions:

• The novel approach demonstrates high accuracy for anatomical landmark classification in upper GI WCE.
• Simulated datasets and advanced image processing techniques can significantly enhance WCE applications in gastroscopy.
• This study contributes to the advancement of WCE technology for broader gastrointestinal examinations.