Identification and reconstruction of bullets from multiple X-rays

We present a framework for the rapid detection and 3D localisation of bullets (or other compact shapes) from a sparse set of cross-sectional patient x-rays. The intention of this work is to assess a software architecture for an application specific alternative to conventional CT which can be leveraged in poor communities using less expensive technology. Of necessity such a system will not provide the diagnostic sophistication of full CT, but in many cases this added complexity may not be required. While a pair of x-rays can provide some 3D positional information to a clinician, such an assessment is qualitative and occluding tissue/bone may lead to an incorrect assessment of the internal location of the bullet.Our system uses a combination of model-based segmentation and CT-like back-projection to arrive at an approximate volume representation of the embedded shape, based on a sequence of x-rays which encompasses the affected area. Depending on the nature of the injury, such a 3D shape approximation may provide sufficient information for surgical intervention.The results of our proof-of-concept study show that, algorithmically, such system is indeed realisable: a 3D reconstruction is possible from a small set of x-rays, with only a small computational load. A combination of real x-rays and simulated 3D data are used to evaluate the technique.