Learning from a tiny dataset of manual annotations: a teacher/student approach for surgical phase recognition (IPCAI 2019)
Tong Yu, Didier Mutter, Jacques Marescaux, Nicolas Padoy
Laparoscopic cholecystectomy is a surgical procedure for removing a patient's gallbladder. As a minimally invasive procedure it is video-monitored via endoscopic cameras.
Our algorithm analyzes the video recordings from those cameras to automatically identify the 7 surgical phases making up the procedure:
- Preparation
- Calot triangle dissection
- Clipping and cutting
- Gallbladder dissection
- Gallbladder retraction
- Cleaning and coagulation
- Gallbladder packaging
The underlying deep neural network is a stack of:
- Resnet-50
- Bidirectional LSTM
- Linear-chain CRF
Training was performed on 80 videos from cholec120, a superset of the publicly released cholec80 dataset available here.
On a test set of 30 videos from cholec120, accuracy reaches 89.5%. Average F1 score over all 7 phases reaches 82.5%.
- Python 3
- Tensorflow 1.14
- numpy
- opencv 3.4
- matplotlib
- ruamel_yaml
Developer configuration info:
- Ubuntu 20.04
- CUDA 10.1
- NVIDIA GTX1080Ti GPU
TF-Cholec80 provides a user-friendly interface for manipulating a dataset of cholecystectomy recordings we previously released. A phase recognition demo using it is available in this repo: (phase_recognition_demo_tfc.ipynb).
@inproceedings{yu2019surgicalphase,
title = {Learning from a tiny dataset of manual annotations: a teacher/student approach for surgical phase recognition},
author = {Tong Yu, Didier Mutter, Jacques Marescaux, Nicolas Padoy},
booktitle = {International Conference on Information Processing in Computer-Assisted Interventions},
year = {2019}
}This code may be used for non-commercial scientific research purposes as defined by Creative Commons 4.0. By downloading and using this code you agree to the terms in the LICENSE. Third-party codes are subject to their respective licenses.