Vertebral pose estimation using horizontal gradient vector field

Automatic and accurate detection of vertebral pose of low resolution x-ray images with different bone structure is essential to diagnose patient condition. Low resolution and incomplete x-ray image is created from low level of irradiating that can avoid radiation over doze. In this research, we propose a novel approach to estimate vertebral pose. There are three main steps including Vertical Projection, Gradient Vector Flow and Feature of Vector Field. Firstly, we segment vertebral zone from full image using Vertical Projection Average with Normal Distribution. Secondly, we introduce Gradient Vector Flow to identify background and foreground of image. Energy calculation helps to divide different objects on image. Then vector field algorithm will help to promote the exact border of the object. Finally, the horizontal field is chosen as the best information to identify the vertebral pose. From the experimental results, we can extract six main bone layouts. There are two normal layouts and four abnormal layouts. For each layout, we deploy different criteria to identify vertebral pose. For the layouts that located between 0-67 index, the separation hyperplane needs to be shifted up, otherwise shifted down. We use 80 low resolution and incomplete x-ray images from a local hospital. The accuracy rate is 79.25% compared with the ground-truth.