SwinCVS: A Unified Approach to Classifying Critical View of Safety Structures in Laparoscopic Cholecystectomy

Authors:
Franciszek Nowak, Evangelos B. Mazomenos, Brian Davidson, Matthew J. Clarkson

---

Overview

Welcome. This repository provides code necessary for reproduction of the SwinCVS publication. The work proposes a SwinV2+LSTM based architecture called SwinCVS, to classify three Critical View of Safety (CVS) criteria from an open access Endoscapes2023 dataset.

Implemented models

• SwinV2 Backbone: Pure SwinV2 backbone. Can be run on random weights or initialised using provided ImageNet weights.
• SwinCSV (E2E, with multiclassifier): SwinCVS with end-to-end training and multiclassifier. Backbone weights initialised on ImageNet.
• SwinCSV (E2E, without multiclassifier): SwinCVS with end-to-end training, but without multiclassifier. Backbone weights initialised on ImageNet.
• SwinCSV (Frozen, without multiclassifier): SwinCVS where image encoding backbone is frozen. Suggested backbone weights pretrained on Endoscapes.

Installation

• Clone this repository
• Confirm you have cuda enabled. In console type nvidia-smi. Our driver API details are: NVIDIA-SMI 550.120 | Driver Version: 550.120 | CUDA Version: 12.4
• Install runtime API cuda 12.1 - Remember to add to path!
• Install dependencies:
  conda create --name swincvs python=3.9.19
  conda activate swincvs
  conda install pytorch==2.4.1 torchvision==0.19.1 torchaudio==2.4.1 pytorch-cuda=12.1 -c pytorch -c nvidia
  pip install -r requirements.txt
  

Usage

Script is run by executing SwinCVS.py from the root of the repository. Specific model training parameters are set within config/SwinCVS_config.yaml. Settings that specify model selection are:

• MODEL.LSTM: False - just SwinV2 backbone training, True - SwinCVS = SwinV2 with LSTM
• MODEL.E2E: False - backbone weights frozen, True - End-to-end training
• MODEL.MULTICLASSIFIER: False - does not add an additional classifier after backbone, True - adds a classifier after backbone, before LSTM
• MODEL.INFERENCE: False - allows for training, True - skips all training, performs only testing on provided weights
• BACKBONE.PRETRAINED: 'str' - which backbone weights to load, imagenet or endoscapes
  

The script automatically downloads the dataset and the weights. If you already have dataset downloaded please specify directory that contains 'endoscapes' folder with all the dataset data.

After changing the config, execute the script by running the SwinCVS.py and specifying which config file to use (default below):
python3 SwinCVS.py --config_path config/SwinCVS_config.yaml

Citation

If you use this work in your research, please cite our paper: Nowak, F., Mazomenos, E., Davidson, B., Clarkson, M., SwinCVS: A Unified Approach to Classifying Critical View of Safety Structures in Laparoscopic Cholecystectomy. Int J CARS (2025). https://doi.org/10.1007/s11548-025-03354-9