Research map / Video / Action recognition
Surgical Workflow Recognition and Blocking Effectiveness Detection in Laparoscopic Liver Resection with Pringle Maneuver
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.
Introduces PmLR50, a 50-video laparoscopic liver resection dataset with Pringle maneuver phase and blocking-effectiveness labels, and PmNet, an online model combining masked temporal encoding, Mamba-based compressed sequence modeling, and contrastive prototype separation.
Problem
Monitoring the Pringle maneuver in laparoscopic liver resection, i.e., recognizing surgical workflow phases and detecting whether intermittent liver inflow blocking is effective, to warn of ineffective or prolonged blocking.
Approach
Combines short-term masked token-swap temporal encoding for fine-grained knotting/releasing actions, a Mamba-based compressed long-term memory that also uses liver ischemia region crops for blocking effectiveness, and a contrastive prototype loss that separates the confusable Knotting and Releasing phases.
Key results
- PmNet reaches 93.83% accuracy and 81.47% Jaccard on PM workflow recognition on the PmLR50 test set, versus 79.10% Jaccard for the second-best Trans-SVNet.
- For blocking effectiveness detection, PmNet reports 98.21% accuracy and 58.33% Jaccard, versus 51.10% Jaccard for Trans-SVNet.
- PmNet runs at 35.34 FPS on a single RTX 4090 with 66.62M parameters, faster than Surgformer (29.85 FPS).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add CPS (ablation) | PmNet · PmLR50 validation, PM workflow recognition | Jaccard: 73.04 -> 74.58 | Helped |
| add CSM (ablation) | PmNet · PmLR50 validation, PM workflow recognition | Jaccard: 71.13 -> 74.58 | Helped |
| add MTE (ablation) | PmNet · PmLR50 validation, PM workflow recognition | Jaccard: 68.14 -> 74.58 | Helped |
| add temporal pooling (ablation) | CSM · PmLR50, blocking effectiveness detection | Jaccard: 58.14 -> 60.87 | Helped |
| add tissue-region box labels (ablation) | CSM · PmLR50, blocking effectiveness detection | Jaccard: 42.11 -> 60.87 | Helped |
| add SSM blocks (ablation) | CSM · PmLR50, blocking effectiveness detection | Jaccard: 35.63 -> 60.87 | Helped |
| add token-swap operation (ablation) | MTE · PmLR50 blocking-related phases | workflow recognition performance | Helped |
| mask blocking-irrelevant features (ablation) | MTE · PmLR50 blocking-related phases | short-term temporal aggregation | Helped |
| use the Knotting-Releasing contrastive pair (ablation) | CPS · PmLR50 PM workflow recognition | Jaccard | Helped |
| use PmNet instead of Trans-SVNet | PmLR50 test, PM workflow recognition · PmLR50 test | Jaccard: 79.10 -> 81.47 | Helped |
| use PmNet instead of Trans-SVNet | PmLR50 test, blocking effectiveness detection · PmLR50 test | Jaccard: 51.10 -> 58.33 | Helped |
Surprising findings
- Ablation on the validation set shows that removing the box (ischemia region) labels raises blocking-detection precision (82.76) but drops Jaccard from 60.87 to 42.11.
Limitations
- Small dataset: 50 procedures, with only 5 ineffective-blocking cases (10%) in total.
- Blocking effectiveness Jaccard remains modest (58.33%), and the gap to baselines on that task is evaluated on a 10-case test set.
Experimental setup
Datasets: PmLR50
Benchmarks: PmLR50
Metrics: Precision Recall Accuracy Jaccard FPS
Baselines: TeCNO TMRNet Trans-SVNet Surgformer
Builds on: Mamba EfficientNet TeCNO Trans-SVNet TMRNet
Closest papers by meaning
- CholecTrack20: A Multi-Perspective Tracking Dataset for Surgical Tools (CVPR 2025 · similarity 0.83)
- EgoExOR: An Ego-Exo-Centric Operating Room Dataset for Surgical Activity Understanding (NeurIPS 2025 · similarity 0.81)
- EndoBench: A Comprehensive Evaluation of Multi-Modal Large Language Models for Endoscopy Analysis (NeurIPS 2025 · similarity 0.80)
- Recognize Any Surgical Object: Unleashing the Power of Weakly-Supervised Data (ICLR 2025 · similarity 0.80)
- SurgicalSAM: Efficient Class Promptable Surgical Instrument Segmentation (AAAI 2024 · similarity 0.80)
- Bridging Vision and Language for Robust Context-Aware Surgical Point Tracking: The VL-SurgPT Dataset and Benchmark (AAAI 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.