基于相位信息的乳腺超声图像水平集分割

发布时间：2018-07-05 17:00

本文选题：超声检查 + 乳腺　；参考：《中国医学影像技术》2015年03期

【摘要】：目的基于相位信息改进距离正规化水平集演化(DRLSE)模型的速度收敛项,改善对乳腺肿瘤超声图像的分割效果。方法首先,利用Log-Gabor滤波器组对图像进行频域滤波,得到一组基于相位信息的特征图。其次,在相位一致性的基础上,求出乳腺超声图像经高斯噪声补偿后的最大方向能量相位PC(M),并采用细节保留各向异性扩散滤波(DPAD)模型对PC(M)降噪,减少斑点噪声的干扰。最后,选用Sigmoid函数,将滤波后的PC(M)作为其自变量,以替换DRLSE模型中的速度收敛项。结果采用改进后的模型对多幅临床乳腺肿瘤超声图像进行分割,分割结果显示基于相位信息的正规化水平集演化(PB-DRLSE)模型在相似性(SI)、真阳性(TP)和假阴性(FN)方面均优于原始DRLSE模型(P均0.05)。结论本研究提出的分割方法较之原始模型对乳腺肿瘤超声图像的分割更为优越。
[Abstract]:Objective to improve the speed convergence term of distance regularization level set evolution (DRLSE) model based on phase information to improve the segmentation effect of breast tumor ultrasound image. Methods first, the Log-Gabor filter bank is used to filter the image in frequency domain, and a set of feature images based on phase information are obtained. Secondly, on the basis of phase consistency, the maximum energy phase PC (M) of breast ultrasound image compensated by Gao Si noise is obtained, and the detail preserving anisotropic diffusion filter model is used to reduce the noise of PC (M) to reduce the interference of speckle noise. Finally, Sigmoid function is used to replace the speed convergence term in DRLSE model by using filtered PC (M) as its independent variable. Results the improved model was used to segment multiple clinical breast tumor images. The segmentation results show that the normalized level set evolution (PB-DRLSE) model based on phase information is superior to the original DRLSE model in similarity (SI), true positive (TP) and false negative (FN) (P 0.05). Conclusion the proposed segmentation method is superior to the original model for breast tumor image segmentation.
【作者单位】：四川大学医学信息工程系;
【基金】：西华大学四川省信号与信息处理重点实验室基金项目(SGXZD0101-10-1)
【分类号】：R445.1;R737.9

【相似文献】