RFMPWC-Net: An Optical flow CNN Network for Ultrasound Elastography

This is the demo code and network weights for

Displacement Estimation in Ultrasound Elastography using Pyramidal Convolutional Neural Network [1].

RFPWC-Net is a CNN network based on PWC-Net [2]. The network is modified for ultrasound elastography. Demo code and fine-tuned weights are available to download. The network architecture is shown in Fig. 1.

Fig.1. RFMPWC-Net from [1]

The network code is based on Pytorch implementation of the PWC-Net and only the sub-network specified in Fig. 1 is fine-tuned and the weights of the other parts are the same as PWC_Net which can be found at https://github.com/NVlabs/PWC-Net/tree/master/PyTorch

Both PWC-Net and the fine-tuned sub-network weights are given here for the comfort of the user. The requirements are the same as Pytorch implementation of PWC-Net and Scipy is also required. Please follow the instruction of PWC-Net to install the toolboxes.

The database used for fine-tuning can be found here:

The estimated displacement using RFMPWC-Net for an experimental phantom is shown below:

The code can be downloaded here:

The demo outputs are shown below:

If you use this demo please cite the following papers:

 [1] A. Tehrani, H. Rivaz, “Displacement Estimation in Ultrasound Elastography using Pyramidal Convolutional Neural Network,” in IEEE Trans. UFFC (TUFFC), Special Issue on Deep Learning, in press

[2] Sun, Deqing, Xiaodong Yang, Ming-Yu Liu, and Jan Kautz. “Pwc-net: Cnns for optical flow using pyramid, warping, and cost volume.” In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 8934-8943. 2018.