运动骨骼肌超声图像序列中肌束方向精细检测与自适应跟踪系统

The estimation of structural parameters of moving skeletal muscles from sequential ultrasound images can facilitate the assessment of locomotor functions and characteristics of skeletal muscles, relevant studies of which have a great significance for a variety of clinical applications. Parameters that can be extracted from musculoskeletal ultrasound images mainly include muscle thickness, fascicle length and fascicle orientation. Since the estimation of muscle thickness and fascicle length in ultrasound images can be achieved based on the detection of fascicle orientation, the detection and tracking of fascicle orientations is the key for the estimation of moving skeletal muscle parameters in musculoskeletal ultrasound image sequence, which is however a challenging task due to the structural complexity and morphological diversity of skeletal muscle during its movement. By conducting more comprehensive experiments in this project, we will complete the detection task of fascicle orientation by developing a new Convolutional Neural Network (CNN) which is able to achieve better estimation accuracy for pixel-wise fascicle orientation estimation in musculoskeletal ultrasound images. Moreover, we will also develop novel adaptive tracking methods based on the developed CNN. By fusing the newly developed CNN detector and adaptive tracker, a system will be established for fine detection and adaptive tracking of fascicle orientations in sequential ultrasound images with moving skeletal muscles. This project is expected to provide experimental tools and technical support for more precise assessment of locomotor functions and characteristics of skeletal muscles, which will finally promote the development of skeletal-muscle-related applications.

超声图像序列中估计运动骨骼肌的结构参数有助于评估骨骼肌运动功能与特性，相关研究具有重要临床应用价值。骨骼肌超声图像中可提取的肌肉参数主要包括：肌肉厚度，肌束长度与肌束方向。其中，肌束长度、肌肉厚度的估计可建立在肌束方向检测基础之上, 因此超声图像序列中估计运动骨骼肌结构参数的关键是肌束方向的检测与跟踪。然而，由于骨骼肌的结构多样性与运动复杂性，超声图像序列中检测并跟踪骨骼肌肌束方向是一项充满挑战性的工作。本项目拟通过较为全面的实验设计，探索与发展适用于运动骨骼肌超声图像肌束方向检测的卷积神经网络，实现对运动骨骼肌超声图像肌束方向更加准确的像素级精细检测，并在此基础上发展自适应跟踪算法，结合卷积神经网络的检测功能，建立一套运动骨骼肌超声图像序列中肌束方向精细检测与自适应跟踪系统，为更加精确地评估骨骼运动功能与特性，促进相关应用发展提供实验平台与技术支持。

超声图像序列中运动骨骼肌结构参数的估计有助于评估骨骼肌运动功能与特性，相关研究具有重要临床应用价值。骨骼肌超声图像中可提取的肌肉参数主要包括：肌肉厚度，肌束长度与肌束方向。其中，肌束长度、肌肉厚度的估计可建立在肌束方向检测基础之上，因此超声图像序列中估计运动骨骼肌结构参数的关键是肌束方向的检测与跟踪。然而，由于骨骼肌的结构多样性与运动复杂性，超声图像序列中检测并跟踪骨骼肌肌束方向是一项充满挑战性的工作。本项目通过发展适用于运动骨骼肌超声图像肌束方向检测的卷积神经网络，实现了对骨骼肌超声图像中肌束方向更加准确的像素级精细检测，并在此基础上发展了自适应跟踪算法，结合卷积神经网络的检测功能，建立了一个运动骨骼肌超声图像序列中肌束方向自适应跟踪系统，为更加精确地评估骨骼运动功能与特性，促进相关应用发展提供了实验平台与技术支持。