Automated recognition of the collimation field in digital radiography images by maximization of the Laplace area integral

In radiographic images the actual region of interest (ROI), i.e. the collimation field, is often smaller than the overall image detector area. Collimation devices (shutters) and lead aprons confine the X-ray beam to the anatomically relevant region. Therefore, large shuttered areas with low radiation intensity may exist in the image. This background may however show strong radiation scatter features, so that simple thresholding or histogram analysis approaches fail. Automated recognition of the collimation field is necessary with respect to optimal contrast adjustment of the monitor and film-printer representation, and accelerates the workflow in comparison to manual ROI settings. In our approach we first identify several hundreds of shutter edge candidates by means of a Hough transform. Then several thousand ROI hypotheses are checked. The objective is to maximize at the same time the enclosed area, the enclosed image intensity, and the enclosed second derivative (Laplace value) of the intensity. The maximization of the Laplace area integral has been found to be the single most powerful feature for finding the true collimation field. The approach was successfully tested on image sets from clinical routine.