神经元网络容错学习在肾小球区域边界增强中的应用

Error Permissibility of Neural Network Used for Renal Corpuscle Area Enhancement

在肾脏组织切片图像的自动分析系统中,肾小球区域边界的增强是一个关键的环节。肾组织切片图像的复杂特点导致了对肾小球边界特征描述的困难。本文在给出特殊边界定义下提出了一种特征模板,用神经元网络构造非线性阈值曲面,考虑网络容错性对边界增强效果的影响,选择合适的阈值曲面进行边界增强。实验结果表明,该容错学习可以在抑制噪声的同时获得对肾小球边界的增强,从而获得良好的处理效果,并且对染色程度不同的样本图像具有强的适应性。

In the automatic analysis system of the kidney-tissue image, boundary enhancement for glomerulus area is a vital step. Complex characteristics of kidney-tissue image leads to the difficulty in boundary features description. This paper suggests a kind of feature template under the special boundary definition. A nonlinear threshold surface is constructed by neural network, then the proper surface can be selected to enhance boundary with the influence of error permissibility being taken into account. Experimental results indicate that this learning method with error permissibility can enhance the boundary of glomerulus and suppress noises at the same time, so it can obtain good processed effects and have a fine performance highly adaptive to various sample images.