LungDepth: Self-Supervised Multi-Frame Monocular Depth Estimation for Bronchoscopy

Area of Science:

• Medical Imaging
• Computer Vision

Background:

• Bronchoscopy is vital for lung biopsies.
• Acquiring depth information from bronchoscopic images enhances procedural intelligence.

Purpose of the Study:

• To develop a self-supervised monocular depth estimation approach for bronchoscopy.
• To improve the accuracy and utility of depth perception in endoscopic procedures.

Main Methods:

• A self-supervised multi-frame monocular depth estimation network was constructed.
• Photometric reprojection error minimization was used for training.
• Adaptive dual attention and details emphasis modules were incorporated to enhance feature capture.

Main Results:

• The proposed method outperformed existing self-supervised monocular depth estimation techniques.
• Superior quantitative and qualitative results were achieved on a custom dataset.

Conclusions:

• The developed monocular depth estimation approach for bronchoscopy demonstrates high accuracy and reliability.
• This advancement supports further research in intelligent bronchoscopic interventions.