Monocular Depth Guided Occlusion-Aware Disparity Refinement via Semi-supervised Learning in Laparoscopic Images

2025-05-13

AI Summary

To address the challenging disparity estimation problem in laparoscopic stereo images—characterized by severe occlusions and scarce annotated data—this paper proposes a Depth-Guided Occlusion-Aware Refinement Network (DGORNet). Methodologically, DGORNet introduces a novel monocular depth prior-guided occlusion modeling paradigm, leveraging depth features to suppress occlusion-induced artifacts; incorporates positional encoding to enhance spatial localization accuracy; and designs an optical flow difference loss (OFD-Loss) to enable semi-supervised spatiotemporal consistency optimization on unlabeled video sequences. On the SCARED dataset, DGORNet achieves significant improvements over state-of-the-art methods in both end-point error (EPE) and root-mean-square error (RMSE), particularly in occluded and textureless regions. Ablation studies confirm the critical contributions of positional encoding and OFD-Loss to robustness in dynamic surgical scenarios.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Search and Optimization: Learning to Search
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Occlusion and the scarcity of labeled surgical data are significant challenges in disparity estimation for stereo laparoscopic images. To address these issues, this study proposes a Depth Guided Occlusion-Aware Disparity Refinement Network (DGORNet), which refines disparity maps by leveraging monocular depth information unaffected by occlusion. A Position Embedding (PE) module is introduced to provide explicit spatial context, enhancing the network's ability to localize and refine features. Furthermore, we introduce an Optical Flow Difference Loss (OFDLoss) for unlabeled data, leveraging temporal continuity across video frames to improve robustness in dynamic surgical scenes. Experiments on the SCARED dataset demonstrate that DGORNet outperforms state-of-the-art methods in terms of End-Point Error (EPE) and Root Mean Squared Error (RMSE), particularly in occlusion and texture-less regions. Ablation studies confirm the contributions of the Position Embedding and Optical Flow Difference Loss, highlighting their roles in improving spatial and temporal consistency. These results underscore DGORNet's effectiveness in enhancing disparity estimation for laparoscopic surgery, offering a practical solution to challenges in disparity estimation and data limitations.

Problem

• Refining disparity maps in laparoscopic images using monocular depth
• Addressing occlusion and limited labeled data in surgical disparity estimation
• Enhancing spatial and temporal consistency in dynamic surgical scenes

Innovation

• Monocular depth guides occlusion-aware disparity refinement
• Position Embedding module enhances spatial context
• Optical Flow Difference Loss improves temporal consistency