基于体动大数据的ADHD分型探索及其脑影像学机制研究

Hyperactivity/impulsivity is one of the core symptoms of Attention Deficit Hyperactivity disorder (ADHD). At present, the evaluation of this symptom is mostly based on the medical history provided by the patient's family members, which lacks objectivity and cannot truly reflect the hyperactivity/impulse symptoms of the child.In this study, ADHD children with attention deficit and hyperactivity/impulsivity were included as well as the normal control group.Kinect was used to record the real-time body movement data of the subjects in the simulated classroom. The method of deep learning was used for analysis, and the body movement model of normal children and ADHD children was established. Through the algorithm of classification and clustering, ADHD was divided into two types, namely adhd-hyperactivity type and adhd-non-hyperactivity type based on objective body movement big data.Then ALFF study of resting state fMRI was conducted on these two types of ADHD children to analyze the brain imaging characteristics of different types of ADHD children, and then explore the possible neural mechanism of hyperactivity/impulsivity symptoms in ADHD children.This study is the first to introduce the artificial intelligence method into the behavioral assessment of ADHD, which provides a basis for exploring the objective indicators of ADHD diagnosis.At the same time, the study combined with the emerging ALFF method provides a new perspective and thought for revealing the hyperactivity/impulsivity mechanism of ADHD.

多动/冲动症状是注意缺陷多动障碍的核心症状之一。目前该症状的评估多根据患者家属提供的病史，缺乏客观性，不能真实的反映患儿的多动/冲动症状。本研究纳入同时具有注意缺陷和多动/冲动症状的ADHD患儿，及正常对照组。采用体动记录仪（Kinect）记录被试在模拟课堂下的实时体动数据，采用深度学习的方法进行分析，建立正常儿童及ADHD患儿的体动模型，并通过分类聚类的算法，将ADHD分为两型，即基于客观体动大数据的ADHD-多动型及ADHD-非多动型。再这对这两型ADHD儿童进行静息态fMRI的ALFF研究，分析不同分型ADHD患儿的脑影像学特征，进而探索ADHD患儿多动/冲动症状可能的神经机制。该研究首次将人工智能的方法引入ADHD的行为学评估，为探索ADHD诊断的客观指标提供了依据。同时结合新兴的ALFF方法进行研究，为揭示ADHD多动/冲动的机制提供了新的视角和思路。

多动/冲动症状是注意缺陷多动障碍的核心症状之一。目前该症状的评估多根据患者家属提供的病史，缺乏客观性，不能真实的反映患儿的多动/冲动症状。本研究纳入同时具有注意缺陷和多动/冲动症状的ADHD患儿，及正常对照组。采用体动记录仪（Kinect）记录被试在模拟课堂下的实时体动数据，采用深度学习的方法进行分析，建立正常儿童及ADHD患儿的体动模型，并通过分类聚类的算法，将ADHD分为两型，即基于客观体动大数据的ADHD-多动型及ADHD-非多动型。再这对这两型ADHD儿童进行静息态fMRI的ALFF研究，分析不同分型ADHD患儿的脑影像学特征，进而探索ADHD患儿多动/冲动症状可能的神经机制。该研究首次将人工智能的方法引入ADHD的行为学评估，为探索ADHD诊断的客观指标提供了依据。同时结合新兴的ALFF方法进行研究，为揭示ADHD多动/冲动的机制提供了新的视角和思路。针对多动儿童目标区域，提出基于噪声位置及灰度分布信息的去噪算法，并用等值线提取多动儿童目标区域的边缘信息，分割出完整的多动儿童身体。然后提出基于 CoM(Center-of-Mass)的运动时长特征，对多动儿童随时间的运动进行量化。实验对多动儿童的视频数据进行测试，结果表明，该算法对多动儿童身体区域的分割正确率为 82.73%~93.77%，运动量化正确率为88.37%~92.47%。