Localizing 2D Ultrasound Probe from Ultrasound Image Sequences Using Deep Learning for Volume Reconstruction
Localizing 2D Ultrasound Probe from Ultrasound Image Sequences Using Deep Learning for Volume Reconstruction
复制标题
DOI：
10.1007/978-3-030-60334-2_10
复制
发表时间：
2020-10
期刊：
影响因子：
--
通讯作者：
Kanta Miura;Koichi Ito;T. Aoki;J. Ohmiya;S. Kondo
中科院分区：
文献类型：
--
作者：
Kanta Miura;Koichi Ito;T. Aoki;J. Ohmiya;S. Kondo
提出了一种基于深度学习的仅从超声图像序列重建超声体积的方法。该方法利用卷积神经网络(CNN)仅从超声图像中估计二维超声探头的位置。我们的CNN模型由两个网络组成：特征提取和运动估计。我们还引入了一致性损失函数来强制执行。通过使用运动捕获系统测量的具有地面真实运动的美国图像序列数据集的一系列实验，我们证明了该方法与传统方法相比在探头定位和体积重建方面表现出了更高的性能。
This paper presents an ultrasound (US) volume reconstruction method only from US image sequences using deep learning. The proposed method employs the convolutional neural network (CNN) to estimate the position of a 2D US probe only from US images. Our CNN model consists of two networks: feature extraction and motion estimation. We also introduce the consistency loss function to enforce. Through a set of experiments using US image sequence datasets with ground-truth motion measured by a motion capture system, we demonstrate that the proposed method exhibits the efficient performance on probe localization and volume reconstruction compared with the conventional method.