Automated multiple lesion identification on vertebral spine using modified average intensity

摘要

X-ray images of vertebral spine are important to analyze vertebral diseases such as osteoporosis, osteopenia and scoliosis. Vertebral Spine lesion affects the diagnosing of bone disease such that the lesion might affect bone mineral density calculation that used to predict osteoporosis. In this respect, the accuracy of identifying lesion from these images is essential. However, some lesion detecting may have low accuracy because of the quality of the acquired images. In this research, we propose a new approach to identify lesion and its position. In our proposed method, there are five main steps. Firstly, Auto Crop, we segment vertebral spine area using Vertical Projection and Normal Distribution. Secondly, Boundary Segmentation, we use Gamma Correction and Distance Regularized Level Set Evolution to detect vertebral spine boundary. Thirdly, Vertebral Pose Estimation, we identify vertebral pose by applying the technique on locating the middle point of a couple of graph peaks as the border line to divide among each pose. Fourthly, Locating Thoracic Bone 12 and Lumbar Bone 5, we use linear equation to localize position of colliding between left or right rib and the vertebral spine to identify Thoracic Bone 12. Likewise, we detect vertebral bone Lumbar Bone 5 by applying the technique to find the minimum distance between position of top hip and vertebral poses. Finally, Identifying Bone Lesion, we use average intensity. From the experimental results, we found that the accuracy performance of our approach is 82.50% for Thoracic Bone 12 and 76.25% for Lumbar Bone 5 compared to ground truth. Moreover, the accuracy performance of bone lesion 61.25%.