基于自适应分数阶微分和改进分水岭的HIFU治疗区域提取方法

Segmentation Method of HIFU Treatment Region Based on Adaptive Fractional Differential and Improved Watershed

在高强度聚焦超声治疗中,需要在超声影像中对目标治疗区域定位与分割。受B超图像复杂背景与噪声干扰,传统的图像分割算法存在局限性。结合Grünwald-Letnikov(G-L)分数阶微分与分水岭思想提出了一种新的图像分割方法。首先将5×5分数阶微分算子划分为八个子模板进行卷积运算,并以归一化后的局部图像梯度与对比度来表征局部图像频率的变化,由此构建自适应分数阶阶次模型,以实现图像的自适应增强;然后对治疗前后的B超图像进行减影,利用形态学开闭操作、距离变换对减影后的图像进行灰度重建,将重建的结果作为分水岭变换的前景标记,以治疗后的B超图像作为背景进行分水岭分割。结果表明:使用该方法分割后的治疗区域轮廓Dice系数,准确率(accuracy)和精度(precision)分别为93.69%、96.74%、92.23%,相较于传统分水岭分割方法分别增加了6.37%、8.59%、6.72%。

In the high-intensity focused ultrasound therapy,it is necessary to locate and segment the target therapeutic region in the ultrasound image.However,due to the complex background and noise of the B-ultrasound image,traditional image segmentation algorithm is limitations.A new image segmentation method is proposed based on Grünwald-Letnikov(G-L)fractional differentiation and watershed idea.Firstly,the 5×5 fractional differential operator is divided into eight sub-templates for convolution operation.Besides,the normalized local image gradient and contrast are used to represent the change of local image frequency from which the adaptive fractional order model is constructed to realize the adaptive enhancement of images.Then,the B-ultrasound images before and after the treatment are subtracted,and the gray scale of the subtraction images is reconstructed by morphological open and close operation and distance transformation.The results of the reconstruction are used as the foreground marker of watershed transformation,and the B-ultrasound images after the treatment function as the background for watershed segmentation.The results indicate that the Dice coefficient,the accuracy and the precision of the treatment area contour segmented by using the proposed method are 93.69%,96.74%and 92.23%respectively,increasing by 6.37%,8.59%and 6.72%when compared with those of the traditional watershed segmentation method.