基于卷积神经网络的注意缺陷多动障碍分类研究

发布时间：2018-04-11 06:44

本文选题：注意缺陷多动障碍 + 磁共振图像　；参考：《生物医学工程学杂志》2017年01期

【摘要】：注意缺陷多动障碍(ADHD)是一种高发于学龄儿童的行为障碍综合症。目前,ADHD的诊断主要依赖主观方法,导致漏诊率和误诊率较高。基于此,本文提出一种基于卷积神经网络的ADHD客观分类算法。首先,对脑部磁共振图像(MRI)进行头骨剥离、高斯核平滑等预处理;其次,对大脑的右侧尾状核、左侧楔前叶和左侧额上回部位的MRI进行粗分割;最后,利用3层卷积神经网络进行分类。实验结果表明:1本文的算法能有效地对ADHD和正常人群进行分类;2右侧尾状核和左侧楔前叶的ADHD分类准确率要高于ADHD-200全球竞赛中所有方法达到的ADHD最高分类准确率(62.52%);3利用上述3个脑区对ADHD患者和正常人群进行分类,其中右侧尾状核的分类准确率最高。综上所述,本文提出了一种利用粗分割和深度学习对ADHD患者和正常人群进行分类的方法。本文方法分类准确率高,计算量小,能较好地提取不明显的图像特征,改善了传统MRI脑区精确分割耗时长及复杂度高的缺点,为ADHD的诊断提供了一种可参照的客观方法。
[Abstract]:Attention deficit hyperactivity disorder (ADHD) is a behavioral disorder with high incidence in school age children.At present, the diagnosis of ADHD mainly depends on subjective method, which leads to high rate of missed diagnosis and misdiagnosis.Based on this, a ADHD objective classification algorithm based on convolution neural network is proposed.First, cranial dissection and smooth nucleus Gao Si were performed on MRI images of brain; secondly, MRI in the right caudate nucleus, left precuneiform lobe and left superior frontal gyrus were roughly segmented; finally, the brain was divided into two parts: the right caudate nucleus, the left anterior cuneate lobe and the left superior frontal gyrus.Three-layer convolution neural network is used to classify.The experimental results show that the ADHD classification accuracy of the right caudate nucleus and the left cuneate lobe in the ADHD and normal population can be effectively classified by the algorithm proposed in this paper, which is higher than the highest ADHD classification accuracy achieved by all the methods in the ADHD-200 Global Competition.The three brain regions mentioned above were used to classify the patients with ADHD and the normal population.The classification accuracy of the right caudate nucleus was the highest.To sum up, this paper proposes a method of classifying ADHD patients and normal population by rough segmentation and deep learning.This method has the advantages of high classification accuracy and less computation. It can extract image features that are not obvious, and improve the disadvantages of traditional MRI brain segmentation, such as long time consuming and high complexity. It provides a referential objective method for the diagnosis of ADHD.
【作者单位】：南昌大学信息工程学院;
【基金】：国家自然科学基金(61463035) 中国博士后科学基金(2016M592117) 江西省科技厅科学基金(20161BAB202045,20151BAB213034) 江西省博士后科研择优项目(2016KY01) 江西省研究生创新专项基金(YC2016-S067)
【分类号】：TP183;R749.94

【参考文献】