Ultrasound motion tracking is required for various medical applications. In this report, we describe a combined approach for tracking anatomical landmarks in liver during respiration, based on the CLUST Challenge. The proposed method combines a local Siamese-CNN and a Ridge Regression temporal model for feature localization at each frame. The method was developed and fine-tuned via 5-fold across-sequence cross-validation and then evaluated on the CLUST Challenge Test set.
This repository contains the code related to the project.
The report describing the methods and the results can also be found under report.pdf.
In order to run any code of this repository 3 environment variables have to be set:
EXP_PATHthe path to the directory saving the checkpointsDATA_PATHthe path to the training dataTEST_PATHthe path to the testing data.
- To run cross_validation evaluation use
global_tracking.py. Set the parameters you wish to use and the name of the experiment in the parameters dictionary at the end of the file and run. - To train, predict, visualize the predictions simply use the jupyter notebook cells in
train_predict_visualize.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent