TPS-RPM based segmentation of three-dimensional high frequency ultrasound SAFT images for CAD/CAM based tooth digitization

Despite many efforts of improving the tooth impression procedure within a tooth restoration process, the detection of subgingival preparations margins still demands retraction of gingiva prior to the impression taking. A concept for high-frequency ultrasound (HFUS) assisted micro-scanning in the context of CAD/CAM based intra-oral tooth surface digitization has been introduced in the past. In order to reduce the mechatronic complexity, the system is based on a two degree of freedom (DOF) highly dynamic serial kinematic carrying a HFUS focused single element transducer. The limited depth of field and the low sensitivity against angular aligned specular reflectors, which might result in missing surface data, are major challenges of the system. In order to enlarge the depth of field and to close the data gaps, the combination of 2-DOF HFUS tooth micro-scanning and 3D synthetic aperture focusing based on a virtual point source (VSAFT) followed by thin plate spline robust point matching (TPS-RPM) based image segmentation is being investigated in this Study. A human molar prepared for crown insertion was scanned in a water tank using an extra-oral laboratory version of the 2-DOF HFUS micro-scanner (75 MHz). RF-data was collected, filtered and processed with 3D VSAFT. After demodulation to base-band and linear gray-value coding, threshold-based maxima detection was applied in order to separate tooth surface echoes (TSE). Then, missing surface data, e.g. parts of the lateral walls of the teeth, were estimated using a TPS-RPM based 2D segmentation approach whereas the initial deformable model was generated based on the TSE data itself. The so reconstructed surface model was compared to a reference model of the same tooth which was automatically scanned by an extra-oral laser-scanning system. The negative / positive mean deviation between the HFUS reconstructed surface model and reference scan was -58 μm/+25 μm, respectively. The standard deviation was 56 &#- 03BC;m. Although the accuracy is worse if compared to earlier presented results from compounded HFUS volume scans, 3D VSAFT together with TPS-RPM based image segmentation reduces system complexity, costs and scanning time.