基于主动轮廓模型的中耳图像分割

目的使用主动轮廓模型进行中耳图像分割,探讨适合中耳图像分割的方法。方法将传统的梯度外力、梯度加压力及梯度矢量流外力应用于主动轮廓模型,分别对砧骨、锤骨、砧骨和锤骨间的共享狭窄面及鼓膜等图像进行图像分割,比较图像分割的结果,选择收敛速度快、分割结果准确的方法。结果当初始化模型非常贴近边缘时,应用梯度或梯度矢量流外力均获得了较好的分割结果;当初始化模型距离边缘较远时,应用梯度外力分割失败,而应用梯度矢量流外力得到了快速而准确的收敛结果;对于鼓膜图像的分割,应用梯度外力和梯度矢量流外力均易越界,应用梯度加压力得到了较好的分割结果。结论对于简单的图像,梯度外力和梯度矢量流外力均能得到较好的分割结果;对于较复杂及边界凹陷的图像,梯度矢量流外力的分割效果明显好于其他外力;对于纤细图像应用梯度加压力的外力更容易控制模型越界的现象。

SEGMENTATION OF MIDDLE-EAR IMAGES BASED ON ACTIVE CONTOUR MODEL

Objective To set up effective active contour models to segment middle-ear images. Methods Traditional gradient force, gradient plus pressure force and gradient vector flow (GVF) force were used in active contour models to segment malleus, incus, eardrum and some shared surface images. The segmentation results were compared. Results Both the gradient force and the GVF force resulted in satisfactory results when the initial contours were close to the boundary. The GVF force also brought satisfactory results when the initial contours were far from the boundary while the gradient force failed. The gradient plus pressure force brought satisfactory results in the eardrum images segmentation while the gradient force and the GVF force failed. Conclusion Both the gradient force and the GVF force can result in satisfactory results in simple images segmentation. The GVF force does better than the other two forces in the segmentation of complex images with concaves. The gradient plus pressure force does better in preventing the model from crossing the slim images.