Segmentation of Femoral Head from CT after Femoral Neck Fracture

Jan Horacek, Lukas Marsalek, Martin Horak, Josef Pelikan, and Philipp Slusallek 
APR Conference on Machine Vision Applications 2009, MVA 2009
Yokohama, Japan, May 20-22, 2009 



We present a semi-automatic method for femoral head segmentation from CT dataset based on finding an optimal path through a polar transformation of axial slices. The cost function is a combination of corticallis properties (the directional behavior of 3D gradients and their magnitude in 2D slices, where they form typical "channels"). The volume is computed using filling and morphological operations. Since no explicit information about the shape of the femoral head is used, the method is also suitable for many other types of fractures or damaged femoral heads. The implementation was experimental ly validated on the Radiology Department of University Hospital Bulovka in Prague and al lows radiologists to intuitively and accurately estimate the femoral head density in approximately 1 to 3 minutes.