Point-Guided Latent Diffusion Model for Novel View Synthesis in Laparoscopic Liver Surgery

Area of Science:

• Medical Imaging
• Computer Vision
• Surgical Technology

Background:

• Synthesizing novel views in laparoscopic liver surgery is difficult due to occlusions and limited fields of view.
• Existing diffusion-based video synthesis methods struggle with accuracy in complex surgical scenarios.

Purpose of the Study:

• To develop a method for generating high-quality intermediate frames in laparoscopic liver surgery from sparse input.
• To improve the accuracy and coherence of surgical scene reconstruction.

Main Methods:

• Proposed a point-guided latent diffusion model integrating 3D point clouds from dense stereo matching.
• Implemented an adaptive camera trajectory planning strategy using next-best-view algorithms to handle occlusions.
• Introduced a spatial-transformer enhanced decoder to preserve anatomical details and reduce visual artifacts.

Main Results:

• The point-guided latent diffusion model demonstrated superior performance in generating visually coherent and structurally accurate novel views.
• Validated effectiveness on the P2ILF challenge dataset for clinically relevant surgical scene reconstruction.
• Successfully addressed challenges of occlusions, complex anatomical shapes, and limited fields of view.

Conclusions:

• The proposed method significantly enhances the quality of surgical scene reconstruction in laparoscopic liver surgery.
• This approach offers a promising solution for improving visualization and navigation during minimally invasive procedures.
• The integration of diffusion models with geometric cues provides a robust framework for complex surgical video synthesis.