A Multi-Level Set Approach for Bone Segmentation in Lumbar Ultrasound Images

Segmentation of the bone surface in ultrasound images are utilized in several applications of computer-aided surgery. In image processing, the bone segmentation is the challenging task in recent days. Hence, a novel multi-level set approach is developed for bone segmentation process and a new fully automatic algorithm for segmenting bone surface is presented in this work. Segmentation of the bone surface is difficult due to speckle noise in the ultrasound image. Ridgelet transform filtering is utilized to remove the speckle noise, which is present in the bone images. The pixel connectivity between neighborhoods in window sequence of region similarity is represented by the clustering formation. The contour of an image represents the segmented area by a novel log-likelihood multi-level set segmentation algorithm. This proposed work provides the accurate segmentation result and high sensitivity, specificity, and precision as 99.92, 94.621, and 93.15. It achieves accuracy range of about 96.5 which is predominant than other existing approaches. Similarly the sensitivity, specificity, precision and recall attain better range. The performance of the proposed method is higher than other segmentation techniques like EM, OTSU and active contour.