Dempster-Shafer Evidence Theory-Based CV Model for Renal Lesion Segmentation of Medical Ultrasound Images
Dempster-Shafer Evidence Theory-Based CV Model for Renal Lesion Segmentation of Medical Ultrasound Images
复制标题
基于 Dempster-Shafer 证据理论的医学超声图像肾脏病变分割的 CV 模型
DOI：
10.1166/jmihi.2017.2080
复制
发表时间：
2017-06
影响因子：
--
通讯作者：
Yin Chen
中科院分区：
文献类型：
--
作者：
Luying Gui;Xiaoping Yang;Armin B. Cremers;Yin Chen
在医学图像分析中，由于图像中存在噪声、对比度低、纹理复杂和不均匀性等问题，对目标进行鲁棒分割是一个具有挑战性的问题。本文提出了一种改进的变分模型，通过基于强度和纹理特征分离感兴趣区域和背景，提高医学图像的分割效果。Dempster-Shafer证据理论融合了这两个来源的信息，并为目标分割提供了令人信服的证据。由于该方法可以捕获物体与背景之间纹理和强度的差异，因此对于病灶的分割是有效的，即使病灶区域与周围环境具有相似的强度。为了证明该方法的有效性，我们使用了52张低对比度且病变不均匀的肾脏超声图像进行定量评价。与经典CV模型、基于预测理论的新分布度量方法、贝叶斯定理融合信息方法、MICO方法和RSF方法相比，该方法的平均精度为94.7%，平均Dice系数为92.8%，是所有测试方法中最高的，平均绝对表面距离（MAD）和对称平均绝对表面距离（SMAD）分别为0.11和0.13。这比其他方法要低。对比结果表明，该方法的分割结果比其他方法更接近真实情况。
In medical image analysis, because of the noises, low contrast, complex textures and inhomogeneities, robust segmentation for objects is a challenging problem. In this paper, a modified variational model is proposed to improve segmentation effects in medical images by separating the regions of interest (ROIs) and backgrounds based on both intensity and texture features. Information from these two sources are fused by the Dempster-Shafer evidence theory and to provide a compelling evidence for object segmentation. Since it can capture the differences of textures as well as intensities between the object and background, the proposed method is effective on segmentation of lesions, even lesion areas that have similar intensities with their surroundings. To prove the efficiency of the proposed method, 52 renal ultrasound images with low contrast and inhomogeneous lesions are utilized in the quantitative evaluations. Compared with the classical CV model, the method using a new distribution metric based on prediction theory, the method using Bayesian theorem to fuse information, the MICO method and RSF method, the proposed method has a mean precision of 94.7%, a mean Dice coefficient of 92.8%, being the highest among all the tested methods, and the mean absolute surface distance (MAD) and the symmetric mean absolute surface distance (SMAD) of this method are 0.11 and 0.13, respectively, which are lower than those of other methods. Comparison results demonstrate that segmentation results by the proposed method are closer to ground truths than those of other methods.