摘要
借鉴模拟退火思想对三维医学图像的最佳熵函数进行拉伸,构造出改进的遗传算法适应度函数,同时采用精英选择策略保留各代最优个体以加快算法收敛速度,从而提出了一种新的基于多阈值最佳熵的三维医学图像分割遗传搜索算法。根据文中算法得到的阈值,成功实现了三维医学脑部图像中脑白质、脑灰质、脑脊液的分割。实验证明,当种群规模为30的情况下,即可搜索到较好的阈值用于三维分割,运算速度较传统穷尽搜索法更快,且比原简单遗传算法具有更强的稳定性和精确性。
Aim. In our opinion, the two shortcomings in traditional searching algorithms are： （1） the simple genetic algorithm （SGA） is deficient in stability and accuracy, （2） the Complete Search （CS） algorithm is time-consuming. We now propose a GA that we believe is better. In the full paper, we explain our better 3D medical image segmentation algorithm in detail. In this abstract, we just add some pertinent remarks to listing the two topics of explanation. The first topic is： the traditional optimal entropy multi-threshold method. The second topic is： the improvement on the 3D image segmentation searching GA using two thresholds. Its two subtopics are the selection of parameters and operators of the GA （subtopic 2. 1） and the improvement on the SGA （subtopic 2.2）. In topic 2, we, using Simulated Annealing idea for reference, stretch the optimal entropy function of 3D medical images to construct the new fitness function of the improved genetic algorithm that adopts the elitist strategy to accelerate the convergence. Finally, we conducted experiments, whose results are summarized in one table and one figure in the full paper. These results show preliminarily that our better algorithm can： （1） find good thresholds; （2） reduce the time required for searching thresholds by 60% as compared with the traditional CS algorithm; （3） most importantly, as compared with SGA, be more stable and accurate, since the relative error is improved from 9. 229 7 × 10^-4 to 1. 762 1 × 10^-4 and the variance is improved from 195.29 × 10^-6 to 2. 220 7×10^-6.
出处
《西北工业大学学报》
EI
CAS
CSCD
北大核心
2007年第3期442-445,共4页
Journal of Northwestern Polytechnical University
基金
国家博士点基金(20040699015)资助
关键词
三维医学图像分割
遗传算法
最佳熵
多阈值
3D medical image segmentation, genetic algorithm （GA）, optimal entropy, multi-threshold
