Pose Estimation of 2D Ultrasound Probe from Ultrasound Image Sequences Using CNN and RNN

In this paper, we propose an ultrasound (US) probe pose estimation method only from US image sequences using deep learning for volume reconstruction. The proposed method employs the combination of convolutional neural network (CNN) and recurrent neural network (RNN) to estimate the US probe pose in light of the long-term temporal information of US image sequences. The features extracted by CNN are input to RNN to estimate the relative and absolute pose of the US probe. Through a set of experiments using US image sequence datasets with ground-truth pose measured by an optical tracking system, we demonstrate that the proposed method exhibits the efficient performance on US probe pose estimation and volume reconstruction compared with the conventional method.