Preliminary study on the automated skull fracture detection in CT images using black-hat transform.

Linear skull fracture, following head trauma, may reach major blood vessels, such as the middle meningeal artery or sinus venosus, and may cause epidural hematoma. However, hematoma is likely to be missed in the initial interpretation because it spreads only gradually. In addition, the fracture lines that run along the scan slice plane are often missed during initial interpretation. In this study, we develop a novel method for automated detection of the linear skull fracture using head computed tomography (CT) images and conduct a basic evaluation using digital phantom and head phantom that enclose genuine human bones. In the proposed method, the bone region is first extracted using morphological processing of the head CT images. Then, the cranial vault is determined from the CT scout view image. The skull has low-density cancellous bone between the hard two-layer high-density compact bones. Because the fracture lines of compact bones are more clearly recognized as compared to cancellous bones, the bone surface is then extracted by performing three-dimensional (3D) Laplacian filtering. Finally, linear structures are extracted by applying the black-hat transform to the bone surface image. In the experiments, we evaluated the proposed method using digital phantom and CT images of the head phantom. From the experiments using digital phantom, we were able to detect a crack line with a width of 0.35 mm. In the experiments using head phantom, we were able to clearly detect the crack lines in the phantom. These results indicate that our proposed method will be useful for the automated detection of skull fracture in CT images.