Attention Deficit Hyperactivity Disorder (ADHD), which has wide distribution all around the world, seriously impacts on personal and family life all along. It is urgent and necessary to introduce new theory proposed in the current research hotspots of brain cognition to analysis it. Under this background, aimed on ADHD magnetic image big data, this project mainly focus on construction of deep neural network with high efficiency after features calculating and evaluating. What’s more, a novel cognition computing research mode is proposed in this project, i.e. ADHD cognitive experiment is designed under the guidance of effective features and discrimination models and conversely, the experimental results are used during feature extraction and discrimination model construction. The main research contents of this project include: Study of effective features extraction and computing based on ADHD big magnetic data; Construct deep neural network for high-dimension and multimodal ADHD fMRI data; Construct deep neural network with high generalization performance for ADHD fMRI data; Fine ADHD cognitive task experiments design based on intelligent computing; Study of ADHD cognition pervasive computing based on calculation ability and cognitive understanding of ADHD features and models. Technically, the research of this project will proposed a series of deep neural networks for heterogeneous, multi-mode, high dimensional data, which will improve the discrimination precision of ADHD. Theoretically, this project will propose a novel spiral cognitive computing research model, i.e., ADHD big data-> model construction -> experimental design -> cognition survey -> model improvement -> experiment improvement -> cognition survey. We think this new research model will help to understand and master the mechanism of ADHD cognition greatly.
注意力缺陷多动障碍症(ADHD)严重影响个人及家庭生活,在脑认知研究成为全球热点的今天,亟需引入新理论对其进行重点研究。本项目面向ADHD核磁影像大数据,从特征的计算评价出发,研究高维高效ADHD深度神经网络构建;创新性地提出特征与模型指导实验设计、实验结果改进模型构建的ADHD认知研究思路。研究内容包括:ADHD多模核磁影像的有效特征萃取与计算;高维多模态ADHD核磁共振大数据的深度神经网络建模与计算;高泛化ADHD深度神经网络结构学习算法研究;基于ADHD认知评价的功能磁共振认知实验设计研究;计算分析与认知理解联合约束的ADHD普适计算研究。本项目理论上深入研究了基于异构、多模、高维数据的ADHD深度神经网络模型,实现高效高泛化判别与分析;在研究方法上提出了“数据—>模型—>实验设计—>认知归纳—>模型改进—>实验改进—>认知归纳”这一螺旋式上升的研究思路,推进ADHD的计算与认知。
本项目针对多模影像数据,开展了如下方面的研究:首先,针对医学图像的高度空间、时间相关性,开展了基于深度神经网络的配准与分割研究,为后续精准分析提供基础;其次,针对医学图像的小样本性,构建小样本集的图像子空间分析模型,实现少样本图像判别与分析,为精准医疗提供技术支撑;第三,针对医学图像中经常出现的数据不平衡问题,开展基于预适应的迁移学习,为智慧医疗提供基础模型支撑;最后,针对智慧医疗中术前、术中及术后可视化的强烈需求,开展了医学影像三维可视化建模,为远程医疗提供技术积累。. 本项目在《IEEE Transactions on Image Processing》、《IEEE Transactions on Neural Networks and Learning Systems》、《IEEE Journal of Internet of Things》、AAAI2020、ICCV2021等国内外学术刊物和会议上共发表论文(含录用)20多篇。申请并公开发明专利6项,软件著作权2项。培养1名博士生,9名硕士生。项目负责人何良华教授获得2019年度教育部科技进步二等奖1项(排名第19),并入选2018年上海市技术带头人,2020年教育部“长江学者奖励计划”特岗学者项目。
{{i.achievement_title}}
数据更新时间:2023-05-31
基于改进LinkNet的寒旱区遥感图像河流识别方法
四川盆地东部垫江盐盆三叠系海相钾盐成钾有利区圈定:地球物理和地球化学方法综合应用
含饱和非线性的主动悬架系统自适应控制
黄曲霉毒素B1检测与脱毒方法最新研究进展
城市生活垃圾热值的特征变量选择方法及预测建模
面向异质遥感影像变化检测的多模深度学习模型与方法
先兆子痫导致认知功能障碍机制的多模影像学研究
基于深度学习的高分辨率PolSAR影像暗目标判别
基于深度学习的多模态神经影像融合分析与脑疾病诊断