归一化Radon变换和统计测度的超声图像肌束方向自动跟踪

Automatic tracking of muscle fascicle orientation in ultrasound images: a approach based on a normalized Radon transform and statistical measure

超声成像具有非侵入、无辐射的特点,被广泛应用于人体肌束的研究。目前,临床上肌束方向的测量主要采用人工方法,主观性强且耗时严重。本文提出利用图像纹理特征来检测和跟踪肌束方向,利用归一化Radon变换对超声图像进行投影,在所有Radon方向利用分段-方差(segmented-variance)法求取不同截距上的统计变化量,变化最大的方向即为肌束方向。与传统的基于直线特征检测的方法相比,该方法充分利用了肌束并行排列的纹理特征,增强了方向检测的鲁棒性与准确性。实验结果表明,与最新的肌束检测方法相比,该方法的鲁棒性得到提高,对4个超声序列进行连续跟踪,其结果与人工检测结果的相关系数为0.90±0.01,两者的平均绝对误差为0.41°±0.31°。该方法在超声影像学分析、生物力学等领域具有极大的应用潜力。

Ultrasound images have been widely adopted to investigate the human muscle fascicles as a non-invasive method that imparts no radiation.However,manual measurement of fascicle orientation from ultrasound images can be subjective and time-consuming in clinical enviroments.In this paper,a new approach is proposed to detect and track the fascicle orientation automatically by using the texture feature of the fascicle region.The ultrasound images are projected using a normalized Radon transform.For each Radon orientation,the variability of the results of the normalized Radon transform is statistically measured by Segmented-Variance.Compared to traditional methods that directly detect the line feature of the fascicles can be directly detected in this way and the proposed method makes full use of the texture feature of the parallel distribution of the fascicles.The result shows that the robustness of the orientation detection has been enhanced.Four ultrasound sequences are tracked consecutively to evaluate this method.The correlation coefficient of the detection results between the proposed method and manual method is 0.90 ± 0.01,and the average of the absolute difference between these two methods is 0.41° ± 0.31°.We believed that the proposed method has a great potential in ultrasonic image analyzing and the biological mechanics field.