Identification and reconstruction of bullets from multiple X-rays

Reviewer: Brad D. Reid

How can an emergency room rapidly locate bullets or other items, such as shrapnel, from a limited number of cross-sectional x-rays__?__ This potentially life-and-death problem is discussed in detail by the authors. Their software research allows facilities without three-dimensional x-ray technology to approximate a computed tomography result with only a set of conventional calibrated x-rays. Healthcare providers and those interested in maximizing their limited technology tools will find this research especially interesting. The authors provide a very well-organized and readable presentation. A background discussion of research in segmenting structures in x-ray images is provided. The authors' segmentation model is discussed next. Metallic objects create areas of high pixel intensity in an x-ray image. The authors' model is based on this property. Their segmentation algorithm is briefly discussed with references to more detailed presentations. Next, their reconstruction process is explained as a "relatively simple ... reconstruction of the two-dimensional slices of a three-dimensional volume." What results did they achieve__?__ Using 12 medical x-ray images from a larger set of 25, they found that "in nine of the twelve x-rays, our algorithm matched the manual segmentation by 90 percent or more, with an average distance contour of 1.3 pixels or less." There were some anomalies in the size of reconstructed bullets due to rounding errors in the reconstruction algorithm. Overall, the result of this work is very encouraging. The authors comment that there is future work to be done in volume rendering, as well as with the issue of rounding errors. The authors also mention a special application of their work to developing countries. The paper contains ample figures and tables, as well as a comprehensive list of references. I recommend it. Online Computing Reviews Service