ADHD classification by a texture analysis of anatomical brain MRI data.
ADHD classification by a texture analysis of anatomical brain MRI data.
复制标题
DOI：
10.3389/fnsys.2012.00066
复制
发表时间：
2012
影响因子：
3
通讯作者：
Chen JH
中科院分区：
文献类型：
--
作者：
Chang CW;Ho CC;Chen JH
ADHD-200全球竞赛为基于静息态功能MRI（rs-fMRI）和结构MRI数据构建注意缺陷/多动障碍（ADHD）诊断分类器提供了绝佳的机会。在这里，我们介绍了一种简单的方法来分类ADHD的基础上形态信息，而不使用功能数据。我们的测试结果表明，这种方法的准确性是有竞争力的方法的基础上的rs-fMRI数据。我们使用三个正交平面上的各向同性局部二值模式（LBP-TOP）从MR脑部图像中提取特征。随后，支持向量机（SVM）被用来开发基于提取的特征的分类模型。在这项研究中，共有436名男性受试者（210名ADHD和226名对照）进行了分析，以显示该方法的区分能力。为了分析这种方法的特性，我们测试了不同的LBP-TOP功能，从各种包裹和不同的图像分辨率。另外，使用单个脑组织类型的形态信息（即，灰质（GM）、白色物质（WM）和CSF）。我们达到的最高精度是0.6995。发现LBP-TOP使用全脑数据作为输入提供更好的辨别力。具有更高分辨率的数据集可以以更高的精度训练模型。转基因的信息比其他组织类型的信息起着更重要的作用。这些结果和LBP-TOP的特性表明，大部分不同的特征分布来自不同的皮质折叠模式。使用LBP-TOP，我们提供了一个ADHD的分类模型，仅基于解剖信息，这是更容易获得在临床环境中，这是更简单的预处理相比，rs-fMRI数据。
The ADHD-200 Global Competition provides an excellent opportunity for building diagnostic classifiers of Attention-Deficit/Hyperactivity Disorder (ADHD) based on resting-state functional MRI (rs-fMRI) and structural MRI data. Here, we introduce a simple method to classify ADHD based on morphological information without using functional data. Our test results show that the accuracy of this approach is competitive with methods based on rs-fMRI data. We used isotropic local binary patterns on three orthogonal planes (LBP-TOP) to extract features from MR brain images. Subsequently, support vector machines (SVM) were used to develop classification models based on the extracted features. In this study, a total of 436 male subjects (210 with ADHD and 226 controls) were analyzed to show the discriminative power of the method. To analyze the properties of this approach, we tested disparate LBP-TOP features from various parcellations and different image resolutions. Additionally, morphological information using a single brain tissue type (i.e., gray matter (GM), white matter (WM), and CSF) was tested. The highest accuracy we achieved was 0.6995. The LBP-TOP was found to provide better discriminative power using whole-brain data as the input. Datasets with higher resolution can train models with increased accuracy. The information from GM plays a more important role than that of other tissue types. These results and the properties of LBP-TOP suggest that most of the disparate feature distribution comes from different patterns of cortical folding. Using LBP-TOP, we provide an ADHD classification model based only on anatomical information, which is easier to obtain in the clinical environment and which is simpler to preprocess compared with rs-fMRI data.