Ajemba Peter, Durdle Nelson, Hill Doug, Raso James
Electrical and Computer Engineering, University of Alberta, W4-040 ECERF, and Rehabilitation Technology, Glenrose Rehabilitation Hospital, Edmonton, AB, Canada, T6G 2V4.
Med Biol Eng Comput. 2007 Jun;45(6):575-84. doi: 10.1007/s11517-007-0192-z. Epub 2007 May 30.
Analysis of three-dimensional (3D) images of human torsos for torso deformities such as scoliosis requires classifying torso distortion. Assessing torso distortion from 3D images is not trivial as actual torsos are non-symmetric and show an outstanding range of variations leading to high classification errors. As the degree of spinal deformity (and classification of torso shape) influences scoliosis treatment options, the development of more accurate classification procedures is desirable. This paper presents a technique for assessing torso shape and classifying scoliosis into mild, moderate and severe categories using two indices, 'twist' and 'bend', obtained from orthogonally transformed images of the complete torso surface called orthogonal maps. Four transforms (axial line, unfolded cylinder, enclosing cylinder and subtracting cylinder) were used. Blind tests on 361 computer models with known deformation parameter values show 100% classification accuracy. Tests on eight volunteers without scoliosis validated the system and tests on 22 torso images of volunteers with scoliosis showed up to 95.5% classification accuracy. In addition to classifying scoliosis, orthogonal maps present the entire torso in one view and are viable for use in scoliosis clinics for monitoring the progression of scoliosis.
分析人体躯干的三维（3D）图像以检测诸如脊柱侧弯等躯干畸形，需要对躯干变形进行分类。从3D图像评估躯干变形并非易事，因为实际的躯干不对称，且呈现出极大的变化范围，导致分类错误率很高。由于脊柱畸形的程度（以及躯干形状的分类）会影响脊柱侧弯的治疗方案，因此需要开发更准确的分类程序。本文提出了一种技术，该技术使用从完整躯干表面的正交变换图像（称为正交图）获得的“扭曲”和“弯曲”两个指标，来评估躯干形状并将脊柱侧弯分为轻度、中度和重度三类。使用了四种变换（轴线变换、展开圆柱变换、包围圆柱变换和相减圆柱变换）。对361个具有已知变形参数值的计算机模型进行的盲测显示分类准确率为100%。对八名无脊柱侧弯的志愿者进行的测试验证了该系统，对22名有脊柱侧弯的志愿者的躯干图像进行的测试显示分类准确率高达95.5%。除了对脊柱侧弯进行分类外，正交图还能在一个视图中呈现整个躯干，可用于脊柱侧弯诊所监测脊柱侧弯的进展情况。
