Background. The gold standard for diagnosing impulsivity relies on clinical interviews, behavioral questionnaires and rating scales which are highly subjective. Objective. The aim of this study was to develop a functional near infrared spectroscopy (fNIRS) based classification approach for correct identification of impulsive adolescents. Taking into account the multifaceted nature of impulsivity, we propose that combining informative features from clinical, behavioral and neurophysiological domains might better elucidate the neurobiological distinction underlying symptoms of impulsivity. Approach. Hemodynamic and behavioral information was collected from 38 impulsive adolescents and from 33 non-impulsive adolescents during a Stroop task with concurrent fNIRS recordings. Connectivity-based features were computed from the hemodynamic signals and a neural efficiency metric was computed by fusing the behavioral and connectivity-based features. We tested the efficacy of two commonly used supervised machine-learning methods, namely the support vector machines (SVM) and artificial neural networks (ANN) in discriminating impulsive adolescents from their non-impulsive peers when trained with multi-domain features. Wrapper method was adapted to identify the informative biomarkers in each domain. Classification accuracies of each algorithm were computed after 10 runs of a 10-fold cross-validation procedure, conducted for 7 different combinations of the 3-domain feature set. Main results. Both SVM and ANN achieved diagnostic accuracies above 90% when trained with Wrapper-selected clinical, behavioral and fNIRS derived features. SVM performed significantly higher than ANN in terms of the accuracy metric (92.2% and 90.16%, respectively, p = 0.005). Significance. Preliminary findings show the feasibility and applicability of both machine-learning based methods for correct identification of impulsive adolescents when trained with multi-domain data involving clinical interviews, fNIRS based biomarkers and neuropsychiatric test measures. The proposed automated classification approach holds promise for assisting the clinical practice of diagnosing impulsivity and other psychiatric disorders. Our results also pave the path for a computer-aided diagnosis perspective for rating the severity of impulsivity.

背景。诊断冲动的黄金标准依赖于临床访谈、行为问卷和高度主观的评分量表。客观的。本研究的目的是开发一种基于功能性近红外光谱 (fNIRS) 的分类方法，用于正确识别冲动的青少年。考虑到冲动的多面性，我们建议结合临床、行为和神经生理学领域的信息特征可能更好地阐明冲动症状背后的神经生物学区别。方法. 在 Stroop 任务期间，从 38 名冲动青少年和 33 名非冲动青少年中收集了血流动力学和行为信息，同时进行 fNIRS 记录。从血流动力学信号计算基于连接的特征，并通过融合基于行为和连接的特征计算神经效率度量。我们测试了两种常用的监督机器学习方法，即支持向量机 (SVM) 和人工神经网络 (ANN) 在使用多域特征训练时区分冲动青少年和非冲动同龄人的效果。Wrapper 方法适用于识别每个域中的信息生物标志物。在运行 10 次交叉验证程序 10 次后计算每个算法的分类准确度，主要结果。当使用 Wrapper 选择的临床、行为和 fNIRS 衍生特征进行训练时，SVM 和 ANN 的诊断准确率均超过 90%。SVM 在准确度指标方面的表现明显高于 ANN（分别为 92.2% 和 90.16%，p = 0.005）。意义。初步研究结果表明，当使用涉及临床访谈、基于 fNIRS 的生物标志物和神经精神病学测试措施的多领域数据进行训练时，这两种基于机器学习的方法在正确识别冲动青少年方面的可行性和适用性。所提出的自动分类方法有望帮助诊断冲动和其他精神疾病的临床实践。我们的研究结果也为评估冲动严重程度的计算机辅助诊断观点铺平了道路。