Li Yichun, Nair Rajesh, Naqvi Syed Mohsen
Intelligent Sensing and Communications Research Group, Newcastle University, UK.
Cumbria, Northumberland, Tyne and Wear (CNTW), NHS Foundation Trust, UK.
Neurosci Appl. 2024 Oct 10;3:104093. doi: 10.1016/j.nsa.2024.104093. eCollection 2024.
Attention Deficit Hyperactivity Disorder (ADHD) is a highly prevalent human neurobehavioral and neurodevelopmental disorder worldwide. Recently, deep learning-based techniques have been exploited in ADHD detection and diagnosis due to their outstanding performance. However, the majority of these methods relying on fMRI and EEG data suffer from the limitation of requiring expensive equipment and incurring high operational costs. Therefore, inspired by the fact that the symptoms of ADHD may manifest in actions and daily behaviors (as stated in the Medical Statistical Manual of Mental Disorders, Fifth Edition (DSM-V)), we introduce a novel ADHD detection system based on human action recognition. We design a novel hyperactivity test for capturing ADHD features and record a real multimodal ADHD dataset (M-ADHD) for the first time. The proposed system detects ADHD symptoms based on acquired action characters from raw RGB videos from the M-ADHD. Our system outperforms conventional competitors in terms of accuracy and AUC on the real multimodal ADHD dataset. Our proposed method, based on simple, non-wearable sensors, has the advantages of being cost-efficient and easy to operate. It is widely applicable for remote ADHD screening and further applies to understanding, treating, and preventing brain disorders.
注意力缺陷多动障碍（ADHD）是一种在全球范围内高度流行的人类神经行为和神经发育障碍。近年来，基于深度学习的技术因其出色的性能而被用于ADHD的检测和诊断。然而，这些大多数依赖功能磁共振成像（fMRI）和脑电图（EEG）数据的方法存在需要昂贵设备且运营成本高的局限性。因此，受ADHD症状可能在行动和日常行为中表现出来这一事实的启发（如《精神疾病诊断与统计手册》第五版（DSM-V）所述），我们引入了一种基于人类动作识别的新型ADHD检测系统。我们设计了一种新颖的多动测试来捕捉ADHD特征，并首次记录了一个真实的多模态ADHD数据集（M-ADHD）。所提出的系统基于从M-ADHD的原始RGB视频中获取的动作特征来检测ADHD症状。在真实的多模态ADHD数据集上，我们的系统在准确性和曲线下面积（AUC）方面优于传统的竞争对手。我们提出的方法基于简单的非可穿戴传感器，具有成本效益高且易于操作的优点。它广泛适用于远程ADHD筛查，并进一步应用于理解、治疗和预防脑部疾病。
