Semiautomatic Method for Segmenting Pedicles in Vertebral Radiographs

The top and bottom of the pedicles are used as landmarks for 3D stereo radiographic reconstruction of vertebrae. At present the landmark identification is manual that can cause observer variability which in turn reduces the accuracy of 3D stereo radiographic reconstruction. A semiautomatic method for segmenting pedicles from biplanar (PA and Lateral) vertebral radiographs is proposed here. Mathematical morphology is proposed for enhancement and Gradient Vector Flow (GVF) snake model is proposed for pedicle segmentation. The conventional theoretical concept of image enhancement has been extended to the regime of multiscale mathematical morphology. Structuring element in this method is multiscale. Bright and dark features at various scales of radiographs are extracted using multiscale tophat transformation. These multiscale features are combined to reconstruct the final modified image. Therefore the contrast of radiograph is enhanced locally.GVF snake model is used for segmenting the pedicles from the enhanced radiograph. A comparison is made between the segmented radiographs with and without the morphological enhancement. Results demonstrated that the distance between contours manually delineated by the user and those segmented by the proposed algorithm is far less than the distance resulted from the traditional GVF snake without morphological enhancement. The morphological enhancement produces better segmentation results even with noisy and low contrast radiographs. The proposed method enables the automatic landmark identification from the segmented radiographs which will remove observer variability that in turn increases the 3D reconstruction accuracy. Hence the proposed method might be a useful preprocessing tool for 3D stereo radiographic reconstruction.