3D ultrasound reconstruction of spinal images using an improved Olympic Hole-filling method

Abstract

We propose a new Hole-filling algorithm by improving the Olympic operator, and we also apply it to generate the volume in our freehand 3D ultrasound reconstruction of the spine. First, the ultrasound frames and position information are compounded into a 3D volume using the Bin-filling method. Then, the Hole-filling method is used to repair gaps in the volume. The conventional Olympic operator defines the empty voxels by sorting the neighboring voxels, removing the n% of the upper and lower values, and averaging them to attain the value to fill the empty voxels. The empty voxel estimation can be improved by thresholding the range width of its neighboring voxels and adjusting it to the average values. The method is tested on a holemanipulated volume derived from a cropped 3D ultrasound volume of a part of the spine. Our MAE calculation on the proposed technique shows improved result compared to all tested existing methods.

Hole filling with oriented sticks in ultrasound volume reconstruction

“…Although nearest neighbor hole filling is popular, 2,4,7,9,12,14 has few parameters, and is considered relatively fast, 9 it has three notable deficiencies. First, it introduces blurring because it uses a mean over many surrounding voxels.…”

“…The most common distribution method is called nearest neighbor distribution, where each individual pixel intensity is assigned to the spatially nearest voxel. 4,5 That voxel then becomes a filled voxel. Overlapping intensities are usually dealt with by computing an average or maximum intensity.…”

“…Second, since sticks hole filling is intended to improve upon nearest neighbor hole filling, 4,7,9,14 we are interested to see how these methods compare. Therefore, nearest neighbor hole filling (using a cube-shaped kernel) is used as our main baseline for comparison.…”

“…The most common hole filling method is nearest neighbor hole filling, 2,4,7,9,12,14 where a kernel region surrounding the hole is searched for filled voxels. The kernel is a cube-shaped grid of voxels, centered on the hole, with an isotropic width.…”

“…Overlapping intensities are usually dealt with by computing an average or maximum intensity. [4][5][6][7] If there is a voxel that is not assigned at least one pixel intensity, then it is considered to be a hole. Holes can occur when the image sampling is too sparse for the output volume, or when the image sampling is uneven (e.g., when there is a rotation component in transducer movement).…”