Human Bone Localization in Ultrasound Image Using YOLOv3 CNN Architecture

Abstract

Localization of human long bones in ultrasound images has quite complex challenges. This is due to a representation of the reflection of a sound wave emitted by a B-scan sensor. The ultrasound scan does not only display bone specimens, but also contains muscles, soft tissue, and other parts under the skin tissue Therefore we need a system that can automatically recognize bone specimens in ultrasound images. This study implements deep learning based learning systems using the convolutional neural network (CNN) method with YOLOv3. The training results from the network detector with IoU threshold 0.5 can recognize bone objects in mAP <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">@50</sub> , mAP <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">@75</sub> and mAP <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">@50:95</sub> with values of 99.98, 97.68 and 85.67 respectively. And for the results of training the network detector with IoU threshold 0.75 can recognize bone objects in mAP <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">@50</sub> , mAP <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">@75</sub> and mAP <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">@50:95</sub> with values of 99.96, 97.46 and 86.35 respectively.

Ähnliche Arbeiten