A concept for fully automated segmentation of bone in ultrasound imaging

This study proposes a novel concept for the automated and computerised segmentation of ultrasound images of bone based on motion information. Force is applied on the heel region using the ultrasound probe and then removed while recording the video of the bone using ultrasound. The interface between the bone and surrounding tissues is the region that moves with maximum speed. This concept is utilised to determine a map of movement, where speed is the criterion used for the bone segmentation from the surrounding tissues. To achieve that, the image is subdivided into regions of uniform sizes, followed by tracking individual regions in the successive frames of the video using an optical flow algorithm. The average movement speed is calculated for the regions. Then, the regions with the higher speed are identified as bone surfaces. It is given as the initial contour for the Chan–Vese algorithm to achieve smoother bone surfaces. Then, the final output from the Chan–Vese is post-processed using a boundary tracing algorithm to get the last automated bone segmented output. The segmented outcomes are compared against the manually segmented images from the experts to determine the accuracy. Bhattacharyya distances are used to calculate the accuracy of the algorithmic and manual output. The quantitative results from Bhattacharyya distances indicated an excellent overlap between algorithmic and manual works (average ± STDEV Bhattacharyya distance: 0.06285 ± 0.002051). The bone-segmented output from the optical flow algorithm is compared with the model output and the texture-based segmentation method’s output. The work from the motion estimation methods has better segmentation accuracy than the model and texture segmentation methods. The results of this study suggest that this method is the first attempt to segment the heel bone from the ultrasound image using motion information.