Automated detection of pelvic fractures from volumetric CT images

Abstract

Pelvic fractures are a major cause of trauma patient mortality. Detection and management of pelvic injuries is challenging due to myriad injury patterns and associated complications such as hemorrhage and infection. In this paper, we propose an automated method of pelvic fracture detection from volumetric CT images. A coarse-to-fine strategy is adopted where a potential region containing the fracture is identified first using intensity and curvature information. The above region is modeled as a weighted graph and a fracture is modeled as a minimum cut in this graph. A second localizing algorithm models the same fracture as a valley, based on the signs of the mean and Gaussian curvature. The minimum cuts as well as the spatial consistent valleys, in isolation, generate a small number of false positives in addition to the true fracture. A joint decision process based on the volumetric graph cuts and the spatially consistent valleys eliminates the false positives. Experimental results indicate the effectiveness of the proposed scheme.