Detection of hairline mandibular fracture using max-flow min-cut and Kolmogorov-Smirnov distance

This paper addresses the clinically challenging problem of hairline mandibular fracture detection from a sequence of Computed Tomography (CT) images. A hairline fracture of critical clinical importance, can be easily missed due to the absence of sharp surface and contour discontinuities and the presence of intensity inhomogeneity in the CT image, if not scrutinized carefully. In this work, the 2D CT image slices displaying a mandible with hairline fractures are first identified within an input CT image sequence of a fractured craniofacial skeleton. This is achieved via an intensity-based image retrieval scheme using Kolmogorov-Smirnov distance as the measure of similarity and an unbroken mandible as the reference image. Since a hairline fracture is essentially a discontinuity in the bone contour, we model it as a minimum cut in an appropriately weighted flow network. The existing graph cut-based segmentation schemes are enhanced with a novel construction of the flow network, guided by the geometry of the human mandible. The Edmonds-Karp refinement of the classical Ford-Fulkerson algorithm is employed next to obtain a minimum cut, which represents the hairline fracture in the already identified CT image slices. Experimental results demonstrate the effectiveness of the proposed method.