深度学习算法在颈动脉超声图像斑块分割中的应用研究

Research on application of deep learning algorithm in carotid artery ultrasound image plaque segmentation

近年来,随着计算机技术的进步和数据集的大规模化,越来越多的人把计算机视觉技术应用到超声医学图像中。但在超声图像方面却存在低准确度且不稳定产生的模糊、伪影等使现有算法对模糊、噪声图像误判较高。另外由于病例过多,人为的去检测和识别斑块过于繁琐。为了缓解这些问题,提出了采用inception的网络结构方法快速准确地获取高噪声的超声图像的关键特征,并通过数据增强和自适应中值滤波的方法确保了分割的稳定性。最后得到的颈动脉分割图像,更有利于医生去观察和判断病变体的严重程度,给医护人员带来了巨大的便利,实验结果表明,本文的方法具有实用意义。

In recent years,with the advancement of computer technology and the large-scale data set,more and more people apply computer vision technology to ultrasound medical images.However,there are low accuracy and unstable blurs and artifacts in ultrasound images,which make existing algorithms misjudge blur and noise images higher.In addition,due to too many cases,it is too cumbersome to manually detect and identify plaques.In order to alleviate these problems,inception’s network structure method is proposed to quickly and accurately obtain the key features of high-noise ultrasound images,and the stability of segmentation is ensured through data enhancement and adaptive median filtering.The final carotid artery segmentation image is more conducive to doctors to observe and judge the severity of the lesion,and it brings great convenience to medical staff.The experimental results show that the method in this paper has practical significance.