A monocular endoscopic image depth estimation method based on a window-adaptive asymmetric dual-branch Siamese...

Area of Science:

• Medical Imaging
• Computer Vision
• Surgical Technology

Background:

• Minimally invasive surgery relies on endoscopic imaging, which often presents challenges like low texture and uneven illumination.
• These image quality issues can compromise surgical and diagnostic accuracy.
• Deep learning offers potential solutions for enhancing endoscopic image processing.

Purpose of the Study:

• To propose a novel monocular medical endoscopic image depth estimation method.
• To improve the accuracy and robustness of depth estimation in challenging endoscopic environments.
• To enhance feature representation for better surgical guidance and diagnosis.

Main Methods:

• A window-adaptive asymmetric dual-branch Siamese network architecture was developed.
• One branch processes global image information, while the other focuses on local details.
• An improved lightweight Squeeze-and-Excitation (SE) module and a cross-attention feature fusion module were incorporated for enhanced feature interaction and representation.

Main Results:

• The proposed method demonstrated superior performance over existing approaches on multiple medical and non-medical datasets.
• Quantitative metrics (RMSE, AbsRel, FLOPs, running time) confirmed the model's effectiveness.
• Qualitative results showed good organ boundary matching capabilities when compared to CT images.

Conclusions:

• The developed deep learning model significantly enhances depth estimation for medical endoscopic images.
• The method shows promise for improving surgical navigation and diagnostic capabilities in clinical settings.
• The approach offers a robust solution to the limitations of traditional endoscopic imaging.