中文摘要
随着成像等各种辅助诊断技术的发展，越来越多跨学科的研究者们基于不同的数据去探索人类大脑的内部运作及其与神经病学和神经退行性疾病之间的联系，然而利用经典的统计学方法处理这种高维医学图像的效果欠佳。本文主要的研究目标是基于注意力缺陷多动障碍的高维影像数据，注意力缺陷多动障碍(Attention Deficit Hyperactivity Disorder，ADHD)也叫少儿多动症，是最常见的儿童疾病之一，可以持续到个人发育的青春期和成年，症状包括难以集中注意力，难以控制行为和活动过度。近年来，在对该病症的影像数据的研究中，传统的统计方法直接将成像数据向量化后会得到超高的维数，这种向量化会对成像数据结构造成严重的破坏，忽视了数据的结构依赖性，从而损失很多重要的结构信息，因此该类数据的处理对经典统计方法提出了前所未有的挑战。 本文所用的实验数据ADHD-200是注意缺陷多动障碍的核磁共振影像(MRI)数据，基于此数据提出了一种多维度分割的数据预处理方法，结合随机森林分类方法对分割后的数据进行分类，然后再根据提出的多维度集成算法进行分类和目标检测任务。主要完成了提高诊断正确率、病变区域检测和阈值选取的三方面研究问题。全世界研究者对数据ADHD-200的平均诊断正确率为60%左右，该方法最终可以达到75.4%。另外，本文分别在模拟数据和真实数据中均进行了实验并检测到了信号区域。最后，结合实际问题背景，本文给出了阈值的选择方法，来帮助医务研究者选取对研究有利的阈值。
关键词：ADHD,MRI，随机森林，数据处理，分类，目标检测
Localization and Classification of Attention Deficit Hyperactivity Disorder Based on Random Forest
Abstract
With the development of imaging and other auxiliary diagnostic technologies, more and more interdisciplinary researchers have been devoted to exploring the
internal operation of human brain and its relationship with neurology and neurode-
generative diseases based on different data.However,the classical statistical method
is not effective in the treatment of high dimensional medical images.The main goal
of this study is based on high-dimensional image data of attention deficit hyper-
activity disorder(Attention Deficit Hyperactivity Disorder,ADHD),also known as
childhood hyperactivity disorder(ADHD),is one of the most common childhood
diseases that can continue into adolescence and adulthood.Symptoms include diffi-
应力强度因子culty in concentration,difficulty in controlling behavior and hyperactivity.In recent
陆虎自由人years,in the study of the image data of the disease,the traditional statistical meth-
ods directly quantize the imaging data to get ultra-high dimension,which will cause
2010江苏高考英语serious damage to the imaging data structure,ignoring the structural dependence
of the data,resulting in the loss of a lot of important structural information,so
the processing of this kind of data poses an unprecedented challenge to the classical
statistical methods.
The experimental data ADHD-200used in this paper is the nuclear magnetic resonance image(MRI)data of attention deficit hyperactivity disorder.Based on
this data,a data preprocessing method of multi-dimensional segmentation is pro-
posed,which is combined with random forest classification to classify the segmented
data.Then the tasks of classification and target detection are then finished accord-
ing to the proposed multi-dimensional integration algorithm,which mainly com-pletes the improvement of diagnosis accuracy,lesion region detection and threshold selection of three aspects of research.The average correct diagnosis rate of data ADHD-200by researchers around the world is about60%,and this method can reach75.4%eventually.In addition,experiments are carried out in both simulated data and real data and the signal region is detected.Finally,combined with the background of practical problems,this paper gives the selection method of threshold to help medical researchers to select the threshold which is beneficial to the research.
Keywords:ADHD,MRI,Random Forest,Classification,Object Detection
目录
中文摘要......................................II 第一章引言. (1)
1.1研究背景与意义 (1)
1.2国内外研究现状 (3)
1.3研究方法和工具 (5)
1.4研究内容 (5)
第二章数据基本统计分析及预处理 (7)
长镜头理论2.1数据与变量说明 (7)
2.1.1数据来源及说明 (7)
2.1.2数据结构及特点 (8)
2.2多维度切片的数据预处理方法 (10)
传播与文化产业
2.2.1多维度切片方法 (10)
环
2.2.2数据预处理过程 (13)
第三章相关理论基础与评价指标 (15)
3.1随机森林算法 (15)
3.2多维度集成方法 (16)
3.2.1图像数据集的判定 (16)
3.2.2分类算法 (17)
3.2.3阈值选择 (18)
3.2.4目标检测算法 (19)
3.3重要假设 (21)
3.4张量和张量分解 (22)
3.4.1张量的定义 (22)
3.4.2CP分解 (23)
3.5分类的评价指标 (23)
第四章模拟数据与实例分析 (25)
4.1模拟数据 (25)
4.1.1模拟数据过程 (25)
4.1.2结果分析 (25)
4.2实际数据 (27)
4.2.1阈值选择 (27)
4.2.2分类结果比较 (28)
4.2.3目标检测 (30)
第五章总结与展望 (33)
5.1研究总结 (33)
5.2存在的问题和展望 (34)
参考文献 (35)
致谢 (39)
豫公网安备41160202000603
站长QQ:729038198
关于我们
投诉建议