Banerjee Sunetra, Ling Sai Ho, Lyu Juan, Su Steven, Zheng Yong-Ping
Annu Int Conf IEEE Eng Med Biol Soc. 2020 Jul;2020:2039-2042. doi: 10.1109/EMBC44109.2020.9175673.
Scoliosis is a 3D spinal deformation where the spine takes a lateral curvature, which generates an angle in a coronal plane. For periodic detection of scoliosis, safe and economic imaging modality is needed as continuous exposure to radiative imaging may cause cancer. 3D ultrasound imaging is a cost-effective and radiation-free imaging modality which gives volume projection image. Identification of mid-spine line using manual, semi-automatic and automatic methods have been published. Still, there are some difficulties like variations in human measurement, slow processing of data associated with them. In this paper, we propose an unsupervised ground truth generation and automatic spine curvature segmentation using U- Net. This approach of the application of Convolutional Neural Network on ultrasound spine image, to perform automatic detection of scoliosis, is a novel one.
脊柱侧弯是一种三维脊柱变形，脊柱会出现侧向弯曲，从而在冠状面产生一个角度。为了定期检测脊柱侧弯，需要一种安全且经济的成像方式，因为持续暴露于辐射成像可能会引发癌症。三维超声成像是一种经济高效且无辐射的成像方式，可提供体积投影图像。已经发表了使用手动、半自动和自动方法识别脊柱中线的相关研究。然而，仍然存在一些困难，比如人为测量存在差异，与之相关的数据处理速度较慢。在本文中，我们提出了一种使用U-Net的无监督真值生成和自动脊柱曲率分割方法。这种将卷积神经网络应用于超声脊柱图像以自动检测脊柱侧弯的方法是一种全新的方法。
