Three-Dimensional Model of Vertebrae Bone Deformation from Bi-Plane X-Ray Image

Abstract

Three Dimension model of human spine become an important material in variety of medical routine. Many applications were proposed to reconstruct a 3D model of human spine. By using images form CT scanner or another High dose of X-ray machine, the model of human vertebrae is deformed. Dual-Energy X-ray Absorptiometry Scanner plays an important role in medical diagnosis and clinical routine. Typically, this kind of machine produces two radiography images of human spine consisted of Anteroposterior and Lateral side. However, these two images provided different viewpoints. Thus, in this proposed method consists of three main phase. Firstly, Region of interest of human lumbar spine is proposed. In the second phase, the vertebrae pose is identified for registration. In the final phase, the three dimensional model of human spine was deformed.

In the first phase, a method for segmenting region of interest on lumbar spine using a horizontal projection and geometric triangulation analysis is proposed. The proposed method offered an automated detection of lumber spine that can reduce computational time and complexity in medical diagnosis. The result of the proposed method can help support many applications such as lumbar vertebrae pose segmentation, bone mineral density analysis and vertebral pose deform In the second phase, we proposed a method to automatically estimate human vertebrae pose in lateral-side image by using Multi-theta Orientation Gabor Filter combined with Principle Component Analysis and Geometric Ellipse Shape Analysis. The proposed method offered an automated estimation of human lumbar vertebrae pose that can reduce work load of radiologist, computational time and complexity in various bone-clinical applications. The result of the proposed method can help support many applications such as lumbar vertebrae pose segmentation, bone mineral density analysis and vertebral pose deformation. The proposed method can estimate the vertebrae pose with 86.4% for accuracy, 80.5% of recall, precision for 89.11% and 13.6% of false negative rate. In the final phase, the three-dimensional model template of human vertebrae is used to deform a model of human lumbar vertebrae from L1 to L4. The template file contains vertices and polygon-faces of lumbar shape in three-dimension space. Then, the surface of the model has been created. The result of this phase is evaluated by Mean Square Error of Volume compare with ground-truth. The L1 and L4 vertebral is gain a huge errors. Because of the information from vertebrae pose estimation step, L1 and L4 are located on a very low contrast area and disturbed with obstacles.

-