Critical View of Safety
Computer Vision–Based Detection of the Critical View of Safety (CVS) in Laparoscopic Cholecystectomy Videos
The Critical View of Safety (CVS) is a well-established surgical criterion used during laparoscopic cholecystectomy to minimize the risk of bile duct injury. Achieving CVS requires clear visualization of specific anatomical structures and conditions before clipping or cutting the cystic duct and artery. Failure to correctly establish CVS remains a major cause of severe surgical complications.
The aim of this project is to develop a computer vision system that analyzes laparoscopic cholecystectomy videos to detect whether the Critical View of Safety has been achieved. Using annotated surgical video data, students will design, implement, and evaluate vision-based methods that identify CVS-related anatomical features and assess CVS status at the frame or video segment level.
Relation to the Reference Dataset
This project is inspired by and aligned with the publicly available, expert-annotated laparoscopic cholecystectomy datasets described in the referenced Nature Scientific Data publication, which provides structured annotations related to CVS assessment and supports reproducible research in surgical AI.
Mentor Details:
Prof. Yoav Mintz
Mentor Details:
Requirments:
Students will aim to:
Study the CVS definition and its visual/anatomical requirements
Analyze laparoscopic cholecystectomy video data and annotations
Develop a computer vision pipeline for CVS detection
Apply deep learning–based models (e.g., CNNs, Vision Transformers) to surgical video
Incorporate temporal context to improve robustness and reduce false positives
Evaluate model performance against expert-labeled CVS annotations
Problem Statement
Detecting CVS from laparoscopic video is a challenging computer vision problem due to:
High anatomical variability across patients
Occlusion by instruments, fat, and inflammation
Dynamic camera motion and viewpoint changes
Partial or ambiguous visualization of required CVS criteria
Subjectivity and variability in expert CVS assessment
CVS is not defined by a single structure, but by the simultaneous satisfaction of multiple visual conditions. The problem is to design a vision-based system that can reliably determine CVS achievement, either as a binary decision or a confidence score, from noisy, real-world surgical video.
Project Objectives
Students will aim to:
Study the CVS definition and its visual/anatomical requirements
Analyze laparoscopic cholecystectomy video data and annotations
Develop a computer vision pipeline for CVS detection
Apply deep learning–based models (e.g., CNNs, Vision Transformers) to surgical video
Incorporate temporal context to improve robustness and reduce false positives
Evaluate model performance against expert-labeled CVS annotations
Technical Scope
The project may include one or more of the following components:
Detection and segmentation of key anatomical structures (e.g., cystic duct, cystic artery, gallbladder, liver bed)
Multi-criteria reasoning to assess CVS conditions
Frame-level or clip-level CVS classification
Temporal modeling using 3D CNNs, Transformers, or recurrent networks
Weakly supervised learning using phase-level or surgeon-provided labels
Required Knowledge and Prerequisites
Core Requirements
Familiarity with fundamental computer vision concepts
Experience with convolutional neural networks (CNNs)
Basic understanding of deep learning frameworks (e.g., PyTorch, TensorFlow)
Ability to work with image and video datasets
Recommended Background
Semantic segmentation and object detection
Video-based learning and temporal modeling
Model evaluation metrics (accuracy, F1-score, AUROC, temporal precision)
No prior surgical knowledge is required; CVS criteria and anatomical context will be provided.
Project Difficulty and Expected Level
Vision complexity: High (multi-structure reasoning, occlusion, ambiguity)
Modeling complexity: High (compositional and temporal reasoning)
Domain knowledge: Low
This project is well-suited for:
Teams of 2–4 students
Expected Outcomes
A working computer vision prototype for CVS detection
Quantitative evaluation against expert-labeled CVS annotations
Analysis of ambiguous cases and failure modes
Well-documented codebase and a technical report