Fully automatic spinal scanning and measurement based on multi-source vision information

Abstract

Ultrasound is a radiation-free alternative to X-ray in the clinical evaluation of human spine. Currently proposed spinal robot-assisted scanning systems rely on single-source vision information, which may result in scan path deviation or require human intervention. Additionally, Cobb angle measurement is almost conducted based on 3D reconstruction of spine. In this article, a fully automatic scanning system is proposed based on multi-source vision information. By fusing ultrasound and RGBD information of human back, the position and pose of the robot-held probe can be determined in real-time, ensuring that vertebrae appear in the center of ultrasound images, and the tight coupling between the probe and the human skin. We also develop an automated Cobb angle measurement method based on landmark localization. The in-vivo and phantom experiments demonstrate that the proposed system can enable the probe to track the human spine, and offer measurement results within error range of 5°.