基于捕食者—食饵粒子群算法和单隐层神经网络算法的病脑检测系统

发布时间：2018-04-18 05:40

本文选题：Hu不变矩 + 磁共振　；参考：《南京师范大学》2017年硕士论文

【摘要】：(1)目的:本文首先介绍了研究背景及意义,然后对磁共振(MR)图像诊断的国内外发展现状做了简单的介绍。本文所提出的智能病脑检测系统(SPBD)即是一种计算机智能辅助诊断病理核磁共振图像系统。人工智能算法的研究将有助于提高检测分类的效率和准确率,在病脑检测领域具有十分重要的意义。本文采用神经网络与MR图像相结合的思路。由于神经网络在分类训练中,数据容易陷入局部最优。所以本文采用了一种较新的,非常有效的捕食者-食饵粒子群算法(PP-PSO)来优化神经网络,从而避免了数据易陷入局部最优问题,增强了SPBD系统的对新数据处理、分类的能力,实现了 SPBD系统对病脑检测的高效,高准确率。(2)方法:本文采用DA-160数据样本,采用Hu不变矩(HMI)来提取脑图像特征,Hu不变矩具有平移、旋转、比例不变性,在目标识别、图像匹配、形状分析等领域都有广泛的应用。本文采用单隐层神经网络(SLN)作为分类器。人工神经网络(ANN)通过模仿人脑形象思维构建神经网络,从而实现分布式的信息处理,具有良好的自适应、自组织和很强的自学能力,是数据分类图像识别的有力工具。用HMI提取得到的一系列由七个特征矩组成的矩阵信息输入SLN,经过SLN训练,输出的结果为非0即1的信息(0表示健康大脑图像,1表示病脑图像)。为了使实验不易陷入局部最优解,本文采用了一种基于粒子群算法(PSO)改进的优化算法——捕食者-食饵粒子群优化算法(PP-PSO)来训练SLN的权值。我们将采用五折分层交叉验证(FFSCV)来对数据进行训练,从而保证了对有限数据集进行尽可能多的学习。最后使用分类准确率作为实验优良的评判标准。(3)结果:将实验结果与其他六种较先进的SPBD算法进行比较,通过训练输出结果对比,发现本文的方法,基于捕食者一食饵粒子群算法和单隐层神经网络算法(HMI + SLN + PP-PSO)分类效果最好,对160个数据集进行测试,灵敏度、特征度和准确率分别达到了: 96.00±5.16%,98.57±0.75%和98.25±0.65%。最后比较了 PSO和PP-PSO分别对应的准确率。其中,PSO作为该实验的优化算法准确率达到96.44%。(4)结论:比较发现,HMI + SLN + PP-PSO分类性能最好,实验结果准确率最高。而且,通过实验结果的比较分析能发现HMI + SLN + PP-PSO方法的优势和不足,为SPBD更进一步的研究和优化做了铺垫。
[Abstract]:Objective: this paper first introduces the background and significance of the research, and then briefly introduces the development of MRI imaging diagnosis at home and abroad.The intelligent brain detection system (SPBDD) proposed in this paper is a computerized intelligent diagnostic system for patho-magnetic resonance imaging (MRI).The research of artificial intelligence algorithm will help to improve the efficiency and accuracy of detection and classification, which is of great significance in the field of brain disease detection.In this paper, the idea of combining neural network with Mr image is adopted.Because the neural network in the classification training, the data is easy to fall into the local optimum.So we use a new and very effective predator-prey particle swarm optimization algorithm (PP-PSO) to optimize the neural network, which avoids the data falling into the local optimal problem and enhances the ability of SPBD system to process and classify the new data.The method of high efficiency and high accuracy of SPBD system for detecting diseased brain is realized. In this paper, DA-160 data sample and Hu invariant moment are used to extract the feature of brain image. Hu invariant moment has translation, rotation, scale invariance, and is used in target recognition.Image matching, shape analysis and other fields have been widely used.In this paper, single hidden layer neural network (SLN) is used as classifier.Artificial neural network (Ann) is a powerful tool for data classification and image recognition, which can construct neural network by imitating human brain image thinking, thus realizing distributed information processing, with good self-adaptation, self-organization and strong self-learning ability.A series of matrix information, which is composed of seven characteristic moments, was extracted by HMI. After SLN training, the output result is that the information of non-zero or 1 represents the healthy brain image / 1 to represent the diseased brain image.In order to make the experiment difficult to fall into the local optimal solution, an improved particle swarm optimization algorithm based on particle swarm optimization (PSO), Predator-prey PSO (Predator-Prey PSO), is used to train the weight of SLN.We will use the FFSCV to train the data, so that we can learn as much as possible from the limited data set.Finally, the classification accuracy rate is used as the excellent criterion of the experiment. The results are compared with the other six advanced SPBD algorithms, and the method of this paper is found by comparing the results of the training output with those of the other six advanced SPBD algorithms.Based on predator-prey particle swarm optimization algorithm and single hidden layer neural network algorithm, HMI SLN PP-PSO-based classification is the best. The sensitivity, characteristic and accuracy of 160 data sets are 96.00 卤5.1610 卤0.75% and 98.25 卤0.65%, respectively.Finally, the accuracy of PSO and PP-PSO are compared.Conclusion: the comparison shows that the classification performance of HMI SLN PP-PSO is the best, and the accuracy of experimental results is the highest.Furthermore, the advantages and disadvantages of the HMI SLN PP-PSO method can be found by comparing the experimental results, which pave the way for the further research and optimization of SPBD.
【学位授予单位】：南京师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R741.044;TP18

【参考文献】