<![CDATA[Bone imaging using ultrasound is a safe technique since it does not involve ionizing radiation and non-invasive. However, bone detection and localization to find its region of interest (RoI) is a challenging task because b-mode ultrasound images are characterized by high level of noise and reverberation artifacts. The image quality is user-dependent and the boundary between tissues is blurry, which makes it challenging to interpret images. In this paper, the deep learning approach using Region Proposal Networks was implemented to detect bone’s RoI in b-mode images. The Faster Region-based Convolutional Neural Network model was fine-tuned to detect and determine the bone location in b-mode images automatically. To evaluate the results, in-vivo experiments were carried out using human arm specimens. A total of 1,066 b-mode bone images from six different subjects were used in the training phase and testing phase. The proposed method was successful in determining the bone RoI with the value of the mAP, the accuracy of detection, and the accuracy of localization of 0.87, 98.33%, and 95.99% respectively.]]>
Copyrights © 2019
