Vertebra localization and centroid detection from cervical radiographs

Abstract

Detection and localization of vertebra body plays a significant role in the diagnosis of many spine disorders such as c-spine trauma and their treatment including cervical surgeries. This requires great effort when one is dealing with poor contrast and noisy set of images like cervical radiographs. This paper aims to describe a technique to localize and detect the centroids of c-spine. The presented technique takes the poor contrast cervical radiograph and enhance the contrast of input image. Then edges of enhanced image are detected. Next, a modified generalized Hough Transform (GHT) is applied on manually selected region of interest (ROI) using a mean model of vertebra body as a template image. As a result voted points, localizing c-spine are obtained and fuzzy c means (FCM) clustering is performed on these points to obtain the centroids of five vertebrae (C3 – C7). The presented technique has been tested on 150 cases of publically available database ‘NHANES II’ and achieved 93.76% accuracy.