Ultrasound imaging for human spine: imaging and analysis

The results of ultrasound imaging of the human spine of a volunteer and the derivation of vertebral axial rotation (AR) and vertebral tilt (VT) are presented. The ARs and VTs were derived semi-automatically. In an ultrasound volume, the vertebral parts were enhanced using a length attribute filter, and localized using Robust Automatic Threshold Selection (RATS). Afterwards, a pair of two prominent regions of each vertebra was selected by ultrasound landmark points. Then, the centers of mass of these regions were used to calculate the ARs and VTs. The ARs and VTs were also determined manually based on the set of landmark points. The difference between the two methods determined the accuracy of the semi-automatic method. The overall results can be categorized into a confident, moderate, and less confident region, respectively, T3–T9, L3–L4, and other vertebrae. For the manual landmark determination, the confident region gave an error range of the ARs determination of 0°–2.1°/0.4°–1.2° (intraobserver/interobserver), and in the VTs determination it had the range of 0.3°–3.2°/0.2°–1.9° (intraobserver/interobserver). The difference between the results of semi-automatic method and the manual results was in the range –1.6° to 1.3°. The results confirm that ultrasound imaging is feasible to scan the human spine, and a semi-automatic method to derive ARs and VTs is also feasible.