Research map / Image generation / Autoregressive and latent generation
Redefining the Laparoscopic Spatial Sense: AI-Based Intra- and Postoperative Measurement from Stereoimages
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Builds a stereo-vision laparoscopic measurement prototype (YOLOv8 tool-tip detection, RAFT-Stereo disparity, Poisson surface reconstruction, Dijkstra on-surface distances) and evaluates direct and on-surface measurement accuracy on a 3D-printed phantom and real surgical images.
Problem
Accurate intra- and postoperative measurement of structures such as vessels, resection margins, and bowel lengths in laparoscopic surgery is laborious and error-prone because structures cannot be measured directly and camera depth perception is limited.
Approach
A passive, stereo-camera-based measurement pipeline that works on the existing laparoscopic video stream: surgical tool tips are segmented by YOLOv8 to define measurement endpoints (online) or clicked manually (offline), then RAFT-Stereo disparity is reprojected to 3D, meshed via Poisson reconstruction, and measured as direct or on-surface (graph shortest-path with spline smoothing) distances.
Key results
- RAFT-Stereo offline direct distance MAE below 0.3 mm (0.18, 0.28, 0.24 mm) with max error 0.79 mm, versus SGBM errors of hundreds of mm.
- Online direct measurements with surgical tool tips had MAE 0.76 to 1.52 mm with max 2.59 mm.
- Spline-interpolated offline on-surface MAE ranged 0.24 to 0.47 mm for planar distances, reduced from basic variant errors of 0.83 to 2.05 mm.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace SGBM with RAFT-Stereo for disparity estimation | Laparoscopic measurement method · Playground, offline direct measurements | MAE ± STD (mm): 732.97 ± 500.94 -> 0.18 ± 0.15 | Helped |
| Use online tool-tip selection instead of offline selection | RAFT-Stereo direct measurement · Playground, direct measurements | MAE ± STD (mm): 0.18 ± 0.15 -> 0.98 ± 0.44 | Hurt |
| Add spline interpolation to on-surface path measurements (ablation) | Laparoscopic measurement method · Playground, offline on-surface measurements | MAE ± STD (mm): 0.83 ± 0.98 -> 0.24 ± 0.18; 2.68 ± 1.78 -> 3.26 ± 1.87 | Mixed |
| Add spline interpolation to on-surface path measurements (ablation) | Laparoscopic measurement method · Playground, online on-surface measurements | MAE ± STD (mm): 2.08 ± 1.47 -> 1.63 ± 1.07; 2.22 ± 2.03 -> 2.30 ± 2.02 | Mixed |
Surprising findings
- SGBM disparity produced implausible measurements (hundreds of mm error) on the textureless, poorly lit phantom, so it was dropped.
- The mesh creation step (PSR), not the point cloud, was identified as the main bottleneck for on-surface accuracy.
Limitations
- Prototype uses an Intel RealSense D435 monochrome stereo camera rather than a true laparoscope.
- Online measurements require both tool tips visible in the field of view and are less accurate than offline ones.
- On-surface measurements produce occasional outliers due to PSR mesh imperfections; the method has not been validated with ground truth in real surgery.
Experimental setup
Datasets: Ex-vivo dVRK segmentation dataset (Colleoni et al. 2020) Middlebury Stereo dataset Hamlyn Centre rectified stereo surgical dataset
Benchmarks: Playground CAD phantom measurements da Vinci surgical phantom (qualitative) Hamlyn Centre stereo surgical images (qualitative)
Metrics: mean absolute error standard deviation of error maximum error
Baselines: SGBM
Builds on: RAFT-Stereo RAFT YOLOv8 Poisson Surface Reconstruction SGBM
Closest papers by meaning
- EgoExOR: An Ego-Exo-Centric Operating Room Dataset for Surgical Activity Understanding (NeurIPS 2025 · similarity 0.82)
- CholecTrack20: A Multi-Perspective Tracking Dataset for Surgical Tools (CVPR 2025 · similarity 0.82)
- Intraoperative 2D/3D Image Registration via Differentiable X-ray Rendering (CVPR 2024 · similarity 0.81)
- MM-OR: A Large Multimodal Operating Room Dataset for Semantic Understanding of High-Intensity Surgical Environments (CVPR 2025 · similarity 0.81)
- Bridging Vision and Language for Robust Context-Aware Surgical Point Tracking: The VL-SurgPT Dataset and Benchmark (AAAI 2026 · similarity 0.81)
- SurgVista: Long-Horizon Surgical World Modeling with Plausible Instrument-Tissue Dynamics (NeurIPS 2026 · similarity 0.80)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.