Scoliosis visualization using ultrasound data
MetadataShow full item record
Scoliosis is a disease characterized by spinal deformation which typically appears during adolescence and growth. Regular assessment to monitor the progression of the disease is important to ensure timely intervention. Routine assessment is performed with X-ray but other techniques are used for preliminary screening or advanced assessment. The limitations of various imaging modalities leave a need for a method for producing 3D patient-specific spinal visualizations suitable for scoliosis assessment using ultrasound imaging. This thesis presents a collection of techniques which, taken together, constitute such a method. After exploring the related background material, a method for producing visualizations from ultrasound-accessible skeletal landmarks is presented. The method uses the transverse process locations to deform a healthy-shaped spine model to match patient anatomy. Visualizations were generated and compared to CT to validate the method. Subsequent developments to the ultrasound assessment process were aimed at reducing operator interaction by automatically segmenting the spine from ultrasound, and generating landmarks to use with the visualization method. A bone segmentation method recently integrated into PLUS was used to identify the bone surfaces in ultrasound scans. Then a variation on k-means estimates the landmark locations. Automatically generated landmarks are prone to containing defects, so a Slicer module offering various correction operations was developed. An ultrasound scan was used to produce a visualization with automatically generated, and subsequently repaired, landmarks. Initial results demonstrate the challenges of automatically generating 3D spinal visualizations from ultrasound data. Landmarks are essentially an under-sampling of a segmentation, and degrade results through a loss of information. Furthermore, operator controlled reparation operations reintroduce user interaction. There is still promise in the overall workflow. The landmark-based visualization method has been used in published work, and modular and incremental developments may improve segmentation generation and interpretation.