超声图像血管分割的研究进展

Advances in blood vessel segmentation in ultrasound image

主要阐述超声图像血管分割算法及其评价指标。基于特征提取的经典图像处理算法不能摆脱对人工的依赖,削弱了分割算法的泛化能力;但对于缺乏大样本超声血管图像的研究场景下,充分利用传统且成熟的技术方法却是一种可行的研究办法。基于机器学习的算法提高了分割算法的泛化能力,改善了传统方法的短板;但深度学习技术对数据的依赖性强、可解释性差,其算法的有效性、稳定性还需深入研究。血管分割评价算法的研究极其重要,研究适合超声图像血管分割的客观评价方法也是重要课题之一。总之,传统方法仍然是解决超声图像血管分割的有效方法,传统方法与深度学习技术的紧密结合是未来的发展趋势。

The blood vessel segmentation algorithm for ultrasound image and its evaluation indexes are mainly reviewed.Although the classical image processing algorithms based on feature extraction can’t get rid of the reliance on manual labor, it is a feasible research approach to make full use of the traditional and mature technical methods in the research scenario where there is a lack of large samples of ultrasound blood vessel images. The algorithms based on machine learning improve the generalization ability of segmentation algorithm and overcome the shortcomings of the traditional methods. However, deep learning techniques have strong dependence on data and poor interpretability, and their effectiveness and stability need to be further studied. The study on blood vessel segmentation evaluation algorithms is critical, and finding the objective evaluation methods suitable for segmentation of blood vessels in ultrasound images is one of the important topics. In conclusion, the traditional method is still an effective method to complete the blood vessel segmentation in ultrasound images, and the combination of the traditional method and deep learning techniques is the future development trend.