Research On 3D Reconstruction Methods Of FREEHAND Ultrasound

Compared with CT and MRI, ultrasound imaging is a promising medicaltechnique for its non-invasiveness, realtime performance, flexibility and low costs.3D ultrasound overcomes the disadvantages of conventional 2D ultrasound imagingand allows the clinicians to intuitively view anatomy in 3D, which assists theclinicians to plan scientific interventional operation and monitor the robots duringthe operation, therefore, leads to a much more realible interventional operation.Aiming to the characteristics of freehand ultrasound images, this paper studiesthe key techniques of 3D ultrasound imaging and constructs an integrated freehand3D ultrasound system including an ultrasound device, an electromagnetic trackingdevice, an image acquisition card and a PC. As freehand ultrasound images arearbitrarily located in space, a position sensor is attached to the probe to record itsrealtime postion and orientation data and temporal calibration is carried out todetermine the offset between the position sensor time-stamps and the B-scantime-stamps. Later the position data is interpolated to match B-scans.Spatial calibration is needed to determine the position and the orientaion of theB-scan plane with respect to the sensor mounted on the probe. After analyzingexisted phantoms, a two-layer Z-fiducial phantom with simple construction is madeto perform spatial calibration. The intersections of the phantom and the B-scn planeare treated as representative points and the calibration transformation is solved byleast-squares fitting algorithm.An algorithm based on pixel nearest neighbour without filling holes is utilizedto ensure realtime performance and accuracy of the reconstruction. The mainstrategies of the algorithms are determined as follows: fully scanning the targetobject to decrease the number of the unfilling voxels, designing the grids and voxelsto reduce computation cost, mapping the pixels in the B-scan plane to the worldcoordinates to assign its value to the nearest voxel, and saving the bigger valuewhen more than one pixel is mapped to the same voxel.In order to guarantee the visulization quality and efficiency, based on VTK, raycasting algorithms is adopted to realize volume views and coordinate transformationis used to realize planar views. Finally, a function experiment of this system isperformed with a pig's kidney. Meanwhile, the system accuracy is analyzed and theexperiments show that the 3D accuracy of the proposed system is better than 3.5mm.