Lim Seongyeon, Dong Suh-Yeon, Tran Bach Xuan, Boyer Laurent, Fond Guillaume, Rahmati Masoud, Duy Nguyen Cao, Latkin Carl, Do Huyen, Ho Roger
Department of Information Technology Engineering, Sookmyung Women's University, Seoul 04310, Republic of Korea.
Division of Artificial Intelligence Engineering, Sookmyung Women's University, Seoul 04310, Republic of Korea.
J Affect Disord. 2025 Nov 15;389:119637. doi: 10.1016/j.jad.2025.119637. Epub 2025 Jun 9.
Functional near-infrared spectroscopy (fNIRS) is increasingly used to examine cognitive abilities and support the diagnosis of bipolar disorder (BD). Despite its growing application, research on functional connectivity in BD remains limited, and task-induced functional connectivity remains under-investigated. This study aims to differentiate patients with BD from healthy controls (HCs) based on fNIRS-derived functional connectivity.
We analyzed fNIRS data from 50 patients with BD and 50 HCs during resting state and two cognitive tasks, including the verbal fluency test (VFT) and Stroop color-word test (Stroop). Functional connectivity was quantified using coherence and correlation evaluations, and four network characteristics, such as local efficiency, global efficiency, global clustering coefficient, and average closeness centrality, were extracted. A thresholding method was applied by filtering the top 10 % to 50 % to highlight significant network connections. Group differences were evaluated using t-tests with false discovery rate.
Patients with BD showed significantly altered coherence values, especially between channels 4 and 10 (p < 0.001, t = -5.534, η = 0.562) and lower global clustering coefficients (p < 0.001, t = -5.666, η = 0.578) compared to HCs. The k-nearest neighbor (KNN) classifier using coherence-based network metrics during the VFT achieved the highest classification performance with an accuracy of 0.818, precision of 0.819, recall of 0.824, and an F1-score of 0.817.
Our findings suggest that task-induced functional connectivity, particularly coherence-based metrics derived during the VFT, reflects distinct functional connectivity patterns that differentiate patients with BD from HCs. These findings support the utility of network-based approaches under cognitive task conditions in characterizing functional connectivity alterations associated with BD.
功能近红外光谱技术(fNIRS)越来越多地用于检测认知能力并辅助双相情感障碍(BD)的诊断。尽管其应用日益广泛,但关于BD功能连接性的研究仍然有限,且任务诱发的功能连接性仍未得到充分研究。本研究旨在基于fNIRS衍生的功能连接性将BD患者与健康对照者(HCs)区分开来。
我们分析了50例BD患者和50例HCs在静息状态以及两项认知任务(包括言语流畅性测试(VFT)和Stroop颜色-词语测试(Stroop))期间的fNIRS数据。使用相干性和相关性评估对功能连接性进行量化,并提取四个网络特征,如局部效率、全局效率、全局聚类系数和平均接近中心性。通过筛选前10%至50%的数据应用阈值方法来突出显著的网络连接。使用带有错误发现率的t检验评估组间差异。
与HCs相比,BD患者显示出显著改变的相干值,尤其是通道4和通道10之间(p < 0.001,t = -5.534,η = 0.562),并且全局聚类系数更低(p < 0.001,t = -5.666,η = 0.578)。在VFT期间使用基于相干性的网络指标的k近邻(KNN)分类器实现了最高的分类性能,准确率为0.818,精确率为0.819,召回率为0.824,F1分数为0.817。
我们的研究结果表明,任务诱发的功能连接性,特别是在VFT期间得出的基于相干性的指标,反映了将BD患者与HCs区分开来的独特功能连接模式。这些发现支持了基于网络的方法在认知任务条件下用于表征与BD相关的功能连接改变的效用。