基于改进水平集图像分割方法的乳腺超声病灶提取

ULTRASOUND GALACTOPHORE LESION EXTRACTION BASED ON REVISED LEVEL SET IMAGE SEGMENTATION

准确高效的乳腺超声病灶提取技术具有重要应用价值,但超声图像灰度不均匀、伪影重、噪声强、乳腺病灶区域与周围组织相似度较高等特有属性给自动分割带来很大挑战。RSF模型是一种较为成功的图像分割方法,但对初始轮廓和噪声较敏感,直接用于病灶提取有待改进。针对图像局部分割需求,通过预提取初始分割区域作为水平集的初始条件,有助于提高分割精度;以局部能量为主导,较好地处理灰度不匀的超声特质,增加全局能量项以使零水平集能够更好地定位在弱边界;引入灰度变化率作用以提高轮廓在灰度匀质部分的演化速度。分割实验结果表明,该方法能较为准确地定位乳腺超声病灶,有一定的临床参考价值。

Accurate and efficient ultrasound galactophore lesion segmentation technology has important applied value,but it faces great challenges from intrinsic properties of the ultrasound images including inhomogeneous intensity,severe artifacts,strong noises,and higher similarity between the galactophore lesion area and the normal tissues around it.RSF model is a comparatively successful image segmentation method,but it is sensitive to initial contour and noise,and needs to be revised if applying to lesion extraction directly.Aiming at the demand of partial image segmentation,taking the pre-extracted initial segmentation region for the initial condition of level set will be helpful to improving the segmentation accuracy; dominated with local energy,we deal with well the ultrasound speciality of inhomogeneous intensity,and increase global energy item to locate the zero level set on weak boundary better; we also introduce the function of gray change rate to accelerate the evolution process of contour in homogeneous intensity domain.Experimental results of segmentation show that the proposed method can more accurately locate the ultrasound galactophore lesion,therefore has certain reference value for the clinic.