https://doi.org/10.21203/rs.3.rs-2843354/v1
This is a computer-aided diagnosis (CADx) system that can determine the optimal view for complete supraclavicular block(SCB) in real time. The segmentation network readily assigns three anatomically integral structures : brachial plexus(yellow), subclavian artery(red), and 1st rib(blue). The classification network estimates an optimallity score of the current image for practicing SCB. A score close to 100 simply means that the image is highly adjusted for SCB.
- python == 3.9
- pytorch == 1.11.0
- opencv-python == 4.5.5.64
For the rest, refer to requirements.txt
git clone https://github.com/nistring/UGA.git
pip install -r requirements.txt
Pretrained weights from https://drive.google.com/drive/folders/1is1dVDRL_owmRBxEQD5pQGGXRQxzXu6u?usp=share_link
First, connect an ultrasound machine to PC.
To execute the realtime optimal view detecting application, for example
python realtime.py --cls_arch resnet34 --seg_arch resnet34 --cls_weights [path_to_cls_weights] --seg_weights [path_to_seg_weights]
Press R to manually assign region of interest(ROI).
Press Q to quit the program.
A demo example of realtime application. Automatically, bounding box is found and can visualize probability of optimal-view in SCB and grad-cam or segmentation predictions.
Receiver operating characteristic curve and precision recall curve. A simple pretrained Resnet34 backbone is found to be effective.
