Monocular endoscopy images depth estimation with multi-scale residual fusion

Area of Science:

• Medical Imaging
• Computer Vision
• Surgical Technology

Background:

• Monocular depth estimation is crucial for clinical endoscopy.
• Endoscopic images present challenges like coherent illumination and texture-less surfaces.
• Existing methods struggle with accurate depth perception in these settings.

Purpose of the Study:

• To propose a novel multi-scale residual fusion method for monocular endoscopy depth estimation.
• To overcome challenges posed by endoscopic imaging conditions.
• To enhance depth estimation accuracy for improved surgical guidance.

Main Methods:

• Utilized image frequency domain component space transformation to stabilize illumination.
• Employed an image radiation intensity attenuation model for initial depth map estimation.
• Applied a multi-scale residual fusion optimization technique for refinement.

Main Results:

• Achieved high structural similarity measures (0.94, 0.82, 0.84) on public datasets.
• Demonstrated depth estimation accuracy of 89.3% and 91.2% on different models.
• Effectively captured intricate details in endoscopic images.

Conclusions:

• The proposed method shows significant potential for clinical endoscopy applications.
• Facilitates reliable depth estimation, enhancing surgical procedure quality.
• Robust performance on public datasets validates its effectiveness.