Demo code for Lateral Strain Imaging using Self-supervised and Physically Inspired Constraints in Unsupervised Regularized Elastography accepted in IEEE Transaction in Medical Imaging [1]. The network weights trained on experimental phantom data is provided here.
The network architecture is MPWC-Net++ [2] which is a modified variant of PWC-Net irr [3]. Please follow the installation guide of original implementation of PWC-Net irr.
You should get the followng results from the demo code.
Networks weights trained on simulation data and using Local Normalized Cross Correlation (LNCC) as the data loss is added (file name: sPICTURE_Simulation.pth.tar).
@article{tehrani2022lateral, title={Lateral Strain Imaging using Self-supervised and Physically Inspired Constraints in Unsupervised Regularized Elastography}, author={Tehrani, Ali KZ and Ashikuzzaman, Md and Rivaz, Hassan}, journal={IEEE Transactions on Medical Imaging}, year={2022}, publisher={IEEE} }
@inproceedings{tehrani2021mpwc, title={MPWC-Net++: evolution of optical flow pyramidal convolutional neural network for ultrasound elastography}, author={Tehrani, Ali KZ and Rivaz, Hassan}, booktitle={Medical Imaging 2021: Ultrasonic Imaging and Tomography}, volume={11602}, pages={14--23}, year={2021}, organization={SPIE} }
@inproceedings{hur2019iterative, title={Iterative residual refinement for joint optical flow and occlusion estimation}, author={Hur, Junhwa and Roth, Stefan}, booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition}, pages={5754--5763}, year={2019} }