Author(s)

A. Kumar & N. Durdle

Abstract

This paper presents a novel procedure for creating a 3D torso image from a pair of stereo digital 2D back images. The aim of this procedure is to obtain 3D images that can be used for assessment of external spinal deformities in scoliosis. Scoliosis is a condition characterized by lateral deviation of the spine coupled with rotation of individual vertebra resulting in visible torso asymmetries. The procedure provides clinicians with a cost effective and mobile setup of acquiring 3D images. To improve the registration process, a novel approach combining tree weighted colour based image segmentation and differential geometry was developed. Image reconstruction involved pre-processing, triangulation and texture application to obtain a 3D image. Analysis was performed using human subjects and objects of known dimension. Evaluation of system performance was done against existing stereovision procedures and range scanning systems. The final 3D image was compared to that obtained from the Konica Minolta Vivid 700 laser scanner. Each image was divided into 360 cross sections for evaluation against size and shape. The 3D image reconstructed from this novel procedure was 75–100% accurate when compared against the 3D image from the laser scanner. The results demonstrate that the procedure is a cost effective clinical tool for assessing torso shape and symmetry. Keywords: scoliosis, registration, image reconstruction, belief propagation, differential geometry, disparity map.

Keywords

scoliosis, registration, image reconstruction, belief propagation, differential geometry, disparity map.