Improved Instance Segmentation-based Algorithm for Surgical Instrument Tip Detection

ICTA 2024.

In this paper, we propose an algorithm based on segmentation masks to identify the tip point of a surgical instrument. We tested it on the Endovis 15 Dataset, achieving an accuracy of up to 87%.

Test

First, clone our repository:

git clone https://github.com/ihbkaiser/TrackingToolTipsLab.git

Next, download our data and trained YOLOv9 models here, then organize the folder structure as shown below

TrackingToolTipsLab/
├── assets/
├── data/
│   └── endovis15/
├── src/
│   ├── Models/
│   │   ├── yolo_weights/
│   │   │   ├── yolov9e-best.pt
│   │   │   └── yolov9c-best.pt
│   │   └── ...
│   └── ...
├── README.md
└── requirements.txt

Next, set up a virtual environment and execute the following commands to see the algorithm being evaluated.

cd TrackingToolTipsLab
python -m venv myenv
myenv\Scripts\activate
pip install -r requirements.txt
cd src
python Eval.py

Demo

The yellow points are the tool tips detected by our algorithm, and the green points are the ground truth tips.

We have uploaded a video to visualize our algorithm. You can view them here.