The process of registering ultrasound (US)images to computed tomography (CT)images relies on accurate segmentation of bony structures in US images. However, segmentation of US images often suffers from the presence of speckle noise, clutter, and acoustic shadowing. We propose to improve the US bone segmentation process with a novel Locally Weighted SLSC (LW-SLSC)beamforming method, which is based on the minimization of the total variation of a spatial coherence weighted sum. Application of this beamformer to an ex vivo human vertebra resulted in a 911% contrast-to-noise ratio (CNR)increase in LW-SLSC images (CNR=23.66)when compared to traditional delay-and-sum (DAS)images (CNR=2.34)created from the same channel data. Application to an ex vivo caprine vertebra with surrounding tissue intact similarly resulted in a 55.8% CNR increase in the LW-SLSC images (CNR=2.01)compared to DAS images (CNR=1.29)created from the same channel data. Bone boundaries in the caprine vertebra were segmented from the US and CT images, and the LW-SLSC beamformer enabled approximately 5.5 mm thinner boundary lines than the DAS beamformer when compared to segmentation results based on CT images. Similarly, the location error of boundary lines was also reduced with 70% of the total spatial error within ±1 mm in LW-SLSC images compared to 47% in DAS images. These results demonstrate that LW-SLSC imaging provides improved bone segmentation over traditional DAS imaging, which has promising implications for real-time segmentation of bone boundaries during spinal fusion surgeries and other procedures that may benefit from accurate US-based bone segmentation.