基于帧间相关性的乳腺MRI肿块分割及重建

Segmentation and 3D reconstruction of breast MRI images based on inter-frame correlations

核磁共振成像(MRI)对乳腺MRI中感兴趣区域(肿块区域)的提取和三维可视化显示可以帮助医生对病变进行更准确的诊断。RSF模型适用于边界模糊的目标分割,利用该模型提出一种基于帧间相关性的乳腺MRI肿块分割方法,并利用分割结果对肿块进行面绘制三维显示。首先对MRI序列图确定迭代顺序,选定关键帧并手动给出初始轮廓,利用RSF模型对其进行分割;将分割结果作为迭代的初始轮廓,采用轮廓迭代的方法将上一帧的分割结果作为下一帧分割的初始轮廓,利用RSF模型对整个序列图进行分割。由于乳腺MRI序列图中相邻几帧间图像灰度分布类似,肿块位置和形状相近等帧间相关性,因此轮廓迭代的方式可以克服RSF模型对初始轮廓敏感的缺点。最后将多组参数下获得的分割结果进行最优化筛选,以提高整个序列的分割精度。将该方法应用于20例乳腺MRI序列图,并与手动分割轮廓进行对比,结果证明,该算法对乳腺MRI中肿块分割具有较高的分割精度和自适应性。

MRI （ Magnetic Resonance Imaging） has become one of the primary means of breast cancer＇ s diagnosis. The extraction of the region of interest （ tumor area） and the 3D visual display can help the doctors make more accurate treatment. Taking advantages of RSF （ Region-Scalable Fitting） model which was applied to the segmentation of the object with fuzzy boundaries, a segmentation method for breast MRI tumor was proposed based on inter-frame correlations and the results were used to get the 3D reconstruction. First the iterative sequence of the MRI diagrams was determined, and the key frame was selected, The key frame was segmented by using RSF model with manually given initial outline and the result was taken as the initial outline of the iteration. The method of initial contour iteration was adopted to get the automatic segmentation of breast MRI sequential diagram and the optimized method was used to screen the results under a set of parameters, improving the accuracy of the segmentation for entire sequence. Because the gray distribution of the adjacent inter-frame image is similar, using iteration can overcome the shortcoming that RSF model is sensitive to the initial outline. The proposed method was applied to 20 cases of breast MRI sequential diagrams. The result was compared with the contour of manual segmentation and demonstrated that the method proposed in this paper achieves a better performance in accuracy and robustness.