Vertebral pose segmentation on low radiation image using Convergence Gravity Force

Abstract

Vertebral pose segmentation is an important factor in diagnosing diseases such as osteoporosis, osteopenia and scoliosis. Low radiation X-ray images are often used to diagnose such diseases. This has been done to reduce patients risk exposure of over dose radiation which may cause from a series of treatments. In this respect, it led to a low accuracy in vertebral pose detection. In this paper, we proposed to improve the automate segmentation of low quality image of vertebral pose with a more generalized technique. In the proposed method, there are three main steps. Firstly, in the pre-processing step, Auto Cropped, Multi-Threshold and Canny Edge Detection are applied to find the vertebral bone structure from the original image. Secondly, Feature Analysis and Gravity Force were used to find the region of interest or the area of each pose. Finally, Colormaps, Intensity Diagnosis and Angle Analysis are adopted to segment each vertebral pose from candidate areas retrieved from second step. The experimental results which were compared with ground truth shown that the proposed approach can estimate vertebral pose with Precision at 79.61% and Recall at 77.11%.