Abstract

The performance of medical image segmentation is generally affected by the parameters of the adopted method and noise. To overcome these issues we introduce in this paper a novel segmentation approach of brain MRI using a region based-active contour model and evolutionary algorithm and without performing any pre-processing step. Our main objective is to accurately extract edges, resolve the intensity inhomogeneity problem and overcome manifestations of noise. Chan and Vese model was adopted by introducing a local similarity factor based on Bilateral filter principle (LSFB). The adjustment of our functional energy parameters was achieved using a multi-population genetic algorithm (MPGA) which can display better search performance than serial single population models, in terms of the quality of the solution found, effort and processing time. We selected Brain MRI from Oasis and Brainweb data base with different noise type. The initialization of the active contour was totally random. A comparison of segmentation results with Chan and Vese model and active contour model with a locally computed signed pressure force (SPF) of Akram and his team reveals a clear efficiency of our proposed approach.

中文翻译：

---

基于区域主动轮廓与相似因子和多种群遗传算法的脑磁共振成像混合分割方法

摘要

医学图像分割的性能通常受所采用方法的参数和噪声的影响。为了克服这些问题，我们在本文中介绍了一种新颖的脑部MRI分割方法，该方法使用基于区域的主动轮廓模型和进化算法，而无需执行任何预处理步骤。我们的主要目标是准确地提取边缘，解决强度不均匀性问题并克服噪声的表现。通过引入基于双边滤波器原理（LSFB）的局部相似因子，采用了Chan和Vese模型。我们的功能能量参数的调整是使用多种群遗传算法（MPGA）实现的，在发现的解决方案质量，工作量和处理时间方面，该算法可以显示出比系列单一种群模型更好的搜索性能。我们从Oasis和Brainweb数据库中选择了具有不同噪声类型的Brain MRI。活动轮廓的初始化是完全随机的。将Chan和Vese模型的分割结果与活动轮廓模型与Akram及其团队的局部计算的有符号压力（SPF）进行比较，可以清楚地看出我们提出的方法的效率。