多模态脑影像深度表示的冰毒成瘾度智能量化研究

Meth refers to the “king of drugs” and has the strong addiction and serious social harmfulness. Addiction severity is the most important index for measuring the extent of harm and abstinence treatment; however, how to measure the addiction severity precisely and efficiently is a key scientific issue to be solved. Both us new types of research and related studies are demonstrating that the transformation of brain function and issues are closely related to the addiction severity of meth addictions. Multimodal images of brains, e.g. fMRIs, MRI and DTI, can represent the transformation of brain function and issues of meth addicts from different points of view. Accordingly, this study aims to quantify the addiction severity of meth addictions by leveraging cutting-edge technologies of informational fusion and deep learning to analyze the multimodal brain medical imaging. Firstly, we newly design deep models to extract semantic information of multimodal brain medical images and build a deep unified multimodal medical image fusion model. Secondly, for building intelligence correlation model between the feature of multimodal brain medical images and the addiction severity, we invite specialists to label the addiction severity of patients according to various clinical standards, compare the different regions between meth addictive observation group and control group, and extract the difference information in order to explore the correlation between the indexes of multimodal brain medical images and the addiction severity. Finally, we design a multi-input deep detector to detect the different regions of brain medical images of meth addictions, and then build an intelligent quantitative model of addiction severity to quantify the meth addiction severity. This study will provide a novel evaluate mechanism based on brain medical images for the addiction severity, and it is highly significant and valuable to the research of addiction mechanism and accurate abstinence treatment of meth addiction.

冰毒被称为“毒品之王”，具有极强的成瘾性与社会危害性。成瘾度是对其危害度与戒断治疗效果的重要评判指标，但是如何精确量化成瘾度是一个亟待解决的关键科学问题。我们及诸多新研究发现，冰毒成瘾者脑功能及脑组织变化与成瘾度密切相关，MRI、DTI等多模态脑影像可从不同角度反映出成瘾者脑功能和组织变化。因此本项目拟利用信息融合、深度学习等前沿技术，对冰毒成瘾者的多模态脑影像数据进行自动分析，实现成瘾度智能量化研究。首先，设计深度融合网络，提取多模态脑影像语义特征，建立多模态脑影像深度特征融合统一描述模型；其次，对比分析成瘾组脑影像与正常组的差异性区域，提取差异性区域信息并研究其与冰毒成瘾度的关联性，建立智能关联模型；最后，设计多输入深度检测网络，自动检测脑影像差异性区域，建立冰毒成瘾度智能量化模型。本研究将提供一种基于脑影像学的成瘾度量化评价新机制，对于冰毒成瘾机制研究及精准戒断具有重要的价值和意义。

冰毒是一种新型合成性精神活性物质，具有强烈的中枢兴奋作用，长期滥用会对人体的认知和神经系统造成不可逆损害，并诱发多种精神疾病。成瘾度是对其危害度与戒断治疗效果的重要评判指标，但是如何精确量化成瘾度，进而实现冰毒成瘾患者的自动筛查是一个复杂的科学问题。本项目从MRI、DTI等多模态脑影像出发，利用信息融合、深度学习等前沿技术，重点对冰毒成瘾者自动筛查开展了系统的研究：（1）建立了多模态脑影像特征深度融合的统一描述模型，实现了基于融减自动编码器的统一描述模型和混合偏移轴向自注意力机制的特征融合算法等模型；（2）挖掘了多模态脑影像信息与成瘾度的关联关系，提出了基于功能连接和脑区域属性的特异脑区挖掘算法，以及基于多粒度级联森林的深度关联模型等；（3）研究了成瘾者自动筛查模型，提出了基于多任务多结构和基于深度多尺度多任务学习的成瘾者自动筛查模型等。上述工作提供一种基于脑影像学的成瘾度量化评价新机制，对于冰毒成瘾机制研究及精准戒断具有重要的价值和意义。项目按照预定计划开展了各项研究工作，实现了预定的研究目标。依托本项目，共发表论文26篇，其中CCF A类论文1篇，CCF B类论文5篇，CCF C类论文4篇，SCI、EI收录19篇，申请发明专利16项，获得软件著作权12项。项目研究部分成果获得山东省科技进步奖三等奖1项和青岛市科技进步奖二等奖1项（项目负责人均为第一完成人），培养博士后5名，研究生32名。项目研发的部分核心算法已在山东省戒毒监测治疗所和山东省生建八三医院进行推广和应用，显著提升了医院的成瘾者戒毒康复治疗水平。

内容获取失败，请点击重试