==== Front Sci Rep Sci Rep Scientific Reports 2045-2322 Nature Publishing Group UK London 79053 10.1038/s41598-020-79053-z Article fNIRS-based functional connectivity estimation using semi-metric analysis to study decision making by nursing students and registered nurses Chong Jie Sheng 1 Chan Yee Ling 1 Ebenezer Esther G. M. 2 Chen Hoi Yen 3 Kiguchi Masashi 4 Lu Cheng-Kai 1 Tang Tong Boon tongboon.tang@utp.edu.my 1 1 grid.444487.f0000 0004 0634 0540Centre for Intelligent Signal and Imaging Research, Institute of Health and Analytics, Universiti Teknologi PETRONAS, 32610 Bandar Seri Iskandar, Malaysia 2 grid.440439.e0000 0004 0444 6368Medicine Based Department, Royal College of Medicine Perak, Universiti Kuala Lumpur, 30450 Ipoh, Malaysia 3 grid.440439.e0000 0004 0444 6368Nursing Programme, Royal College of Medicine Perak, Universiti Kuala Lumpur, 30450 Ipoh, Malaysia 4 grid.417547.40000 0004 1763 9564Research & Development Group, Hitachi Ltd., Tokyo, 185-8601 Japan 16 12 2020 16 12 2020 2020 10 220414 5 2020 1 12 2020 © The Author(s) 2020Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.This study aims to investigate the generalizability of the semi-metric analysis of the functional connectivity (FC) for functional near-infrared spectroscopy (fNIRS) by applying it to detect the dichotomy in differential FC under affective and neutral emotional states in nursing students and registered nurses during decision making. The proposed method employs wavelet transform coherence to construct FC networks and explores semi-metric analysis to extract network redundancy features, which has not been considered in conventional fNIRS-based FC analyses. The trials of the proposed method were performed on 19 nursing students and 19 registered nurses via a decision-making task under different emotional states induced by affective and neutral emotional stimuli. The cognitive activities were recorded using fNIRS, and the emotional stimuli were adopted from the International Affective Digitized Sound System (IADS). The induction of emotional effects was validated by heart rate variability (HRV) analysis. The experimental results by the proposed method showed significant difference (FDR-adjusted p = 0.004) in the nursing students’ cognitive FC network under the two different emotional conditions, and the semi-metric percentage (SMP) of the right prefrontal cortex (PFC) was found to be significantly higher than the left PFC (FDR-adjusted p = 0.036). The benchmark method (a typical weighted graph theory analysis) gave no significant results. In essence, the results support that the semi-metric analysis can be generalized and extended to fNIRS-based functional connectivity estimation. Subject terms Problem solvingBiomedical engineeringImaging and sensinghttp://dx.doi.org/10.13039/501100003093Ministry of Higher Education, MalaysiaHICOE to CISIRTang Tong Boon issue-copyright-statement© The Author(s) 2020 ==== Body Introduction Graph theory has been widely employed in neuroimaging studies such as functional magnetic resonance imaging (fMRI), electroencephalographic (EEG) and functional near-infrared spectroscopy (fNIRS) to better understand the functional connectivity (FC) under various neurological conditions1–4. According to the graph theory, a brain network may be described as a graph consisting of nodes, and the connections between the nodes are known as edges5. A collection of nodes in a brain network forms a brain region while the strength or synchronicity of the connectivity is usually represented by the weight of the edges. The number of edges connected to a node is referred to as a degree. To quantitatively evaluate the information transmission ability in a network, the shortest path length plays a crucial role in defining network efficiency at both local and global levels3. In conventional graph theory analysis, the shortest path length is always defined as the minimum sum of the distance between two nodes6, providing the most preferable route for information to be passed from one node to another. In real-world networks, the shortest path is not always the direct distance because the distance from one node to another via a circuitous (indirect) path may be less than the length of the direct path7. This phenomenon violates the transitive property and forms a semi-metric network8, as shown in Fig. 1. Previous studies on fMRI7,9 have shown improvement in graphical network processing performance in information sharing by employing a semi-metric network via indirect paths. For instance, semi-metric networks have succeeded in discriminating changes in the human FC in different neurological conditions based on the percentage of the network semi-metricity8,10,11. However, its reproducibility and generalizability to other neuroimaging modalities such as fNIRS are unclear.Figure 1 An example of a semi-metric network with weighted path. The dashed line represents the direct path from A to C with a weight of 7. However, there exists a shorter circuitous path from A to C via B with a sum of weight of 6. This is called an indirect path. In semi-metric analysis, a network graph consists of two main components: semi-metric edges and metric backbones. Both these components have different shortest path properties. When a preferential (shortest) route is found via a circuitous path, this indirect path between the nodes is called a semi-metric edge. These semi-metric edges depict a higher level of information sharing among multiple brain regions along the circuitous paths, in contrast to the information exchange by metric backbone edges10. This network redundancy feature has not been considered in conventional graph theory-based FC analysis. Such characterization by semi-metric analysis is crucial for describing the hyperconnectivity of the FC. Excessive hyperconnectivity, for instance, has been identified as one of the main symptoms of neurodevelopmental disorder (autism) and post-traumatic stress disorder (emotional processing)12,13. This study proposes the application of fNIRS to measure neuronal activity in the prefrontal cortex (PFC). The fNIRS data was found to closely correlate to that of the fMRI14. Importantly, fNIRS devices are generally less expensive, provide higher portability and better balance between spatial and temporal resolution, and also provide higher resistance to motion artifacts15. Few studies have been reported using fNIRS to assess functional connectivity during the resting state and cognitive task2,16,17. These studies utilized semi-automatic methods for graph filtering, which involved manual tuning of the threshold to select edges above a certain weight value. Such methods may cause inconsistency in the interpretation of FC networks. In this study, we propose to explore a semi-matrix for fNIRS-based FC analysis for the first time. We hypothesize that (i) the semi-metric analysis applied in fMRI FC estimation could be generalized and extended to fNIRS, and (ii) the semi-metric analysis may provide a more effective technique to evaluate fNIRS-based FC than the conventional graph theory approach. To illustrate, we implemented the proposed method to study the differential FC under affective and neutral emotional states in nursing students and registered nurses. Studies have shown that nursing is one of the most stressful jobs18, and the drop-out rate among nursing students is very high (the average drop-out rate was 35–37% for Italy in 2011, 21.1% for Netherlands in 2013, and 20% for UK in 2015)19. The occupational stress results in reduced productivity among nurses (i.e., distracted decision-making ability), threatening patient care. To combat mental stress, registered nurses were found to cope with stress effects by exercising emotion regulation20. This study aims to develop a means to detect the dichotomy in differential FC under affective and neutral emotional states in nursing students and registered nurses as the first step to identifying neural markers among nursing students who may be more susceptible to stress coping issues. Results This experiment was conducted in two sessions for each subject (under affective and neutral emotional states), in a counterbalanced manner. Our statistical analysis showed that the FC, HRV, and behavioral performance results were not affected (p > 0.05) by session order (affective-neutral session versus neutral-affective session). HRV analysis Using two-way mixed analysis of variance (ANOVA), the experimental results of root mean square of successive differences (RMSSD) showed a two-way interaction [F(1,36) = 4.148, p = 0.049, ηp2 = 0.103] between group type and emotional state. In the subsequent simple effect analysis, the result revealed a significant emotional effect [F(1,18) = 8.117, p = 0.011, ηp2 = 0.311]. The pairwise comparison with false discovery rate (FDR) adjustment (in Fig. 2) showed that nursing students possessed significantly lower RMSSD (FDR-adjusted p = 0.044, t(18) =  − 2.834, Cohen’s d = 0.654) in affective state than that in neutral emotional state. On the other hand, no significant difference was observed in RMSSD among registered nurses (p > 0.05).Figure 2 The result of HRV analysis based on RMSSD. * indicates FDR-adjusted p < 0.05. Semi-metric analysis Based on the two-way mixed ANOVA in global level semi-metric analysis, a significant two-way interaction between group type and emotional state was observed [F(1,36) = 5.063, p = 0.031,  ηp2 = 0.123]. Followed by the simple main effect analysis, significant emotional state effect was only observed in the students [F(1,18) = 15.212, p = 0.001,  ηp2 = 0.458]. From the pairwise comparison with FDR correction (in Table 1), the affective state resulted in significantly higher global semi-metric percentage (SMP) (FDR-adjusted p = 0.004, t(18) = 3.922, Cohen’s d = 0.895) than the neutral emotional state among the students. None of the comparison was significantly different in the case of nurses (p > 0.05).Table 1 Pairwise comparison between parameters in semi-metric analysis and graph theory analysis at global level. Comparisons df SMP Eglobal Elocal CC λ Students Affective vs. neutral 18 0.004; 3.922; 0.895* 0.760; 0.900; 0.206 0.766; 0.894; 0.205 0.564; 0.895; 0.205 0.590; 1.079; 0.250 Nurses Affective vs. neutral 18 0.911; 0.113; 0.026 0.760; 1.158; 0.275 0.766; 1.135; 0.275 0.564; 1.136; 0.250 0.590; 1.102; 0.252 Affective Nurses vs. students 36 0.494; 1.178; 0.393 0.874; 0.159; 0.053 0.899; 0.129; 0.043 0.898; 0.129; 0.043 0.912;0.112; 0.037 Neutral Nurses vs. students 36 0.548; − 0.831; 0.277 0.847; 0.385; 0.128 0.899; 0.356; 0.119 0.898; 0.356; 0.119 0.912; 0.204; 0.068 The table are shown in (FDR-adjusted p-values; t-values; Cohen’s d). df indicates degrees of freedom. * represents FDR-adjusted p < 0.05. A three-way mixed ANOVA was carried out to study the interaction between the group type, emotional state, and brain region. The regional semi-metric analysis result indicated a significant three-way interaction effect [F(1,36) = 8.278, p = 0.007,  ηp2 = 0.187]. Based on a follow-up evaluation (by splitting groups into students and nurses), the simple two-way interaction was found to be significant in the students [F(1,18) = 20.889, p < 0.001, ηp2 =  0.537], but not in the nurses [F(1,18) = 0.106, p = 0.749, ηp2 =  0.006]. Result of a follow-up analysis further showed that the students had a significant simple main effect of task [F(1,18) = 13.454, p = 0.002, ηp2 =  0.428] in the right PFC, but not in the left PFC [F(1,18) = 0.016, p = 0.902, ηp2 =  0.001]. The effect of region was also found to be significantly different in the affective state [F(1,18) = 9.740, p = 0.006, ηp2 =  0.351], but not in the neutral emotional state [F(1,18) = 1.072, p = 0.314, ηp2 =  0.056]. Lastly, the multiple pairwise comparison was conducted according to the significant simple main effects. With the FDR correction, Table 2 showed that the SMP in the right PFC was significantly greater (FDR-adjusted p = 0.036, t(18) = 3.113, Cohen’s d = 0.716) than that in the left PFC among the students in the affective state. Moreover, in the right PFC, the SMP was found to be significantly higher in the affective state than that in the neutral emotional state among the students (FDR-adjusted p = 0.024, t(18) = 3.298, Cohen’s d = 0.841). Meanwhile, none of the comparisons were significantly different in the case of the nurses (p > 0.05).Table 2 Pairwise comparison of FC indices in regional semi-metric analysis and graph theory analysis. Comparisons df SMP Enodal CC λ Nurses Affective Right vs. left PFC 18 0.604; 0.923; 0.212 0.907; 0.316; 0.066 0.928; 0.293; 0.057 0.961; 0.196; 0.033 Neutral Right vs. left PFC 18 0.555; 0.919; 0.316 0.907; − 0.326; 0.076 0.928; − 0.332; 0.076 0.948; − 0.382; 0.076 Students Affective Right vs. left PFC 18 0.036; 3.113; 0.716* 0.876; 1.069; 0.258 0.858; 1.029; 0.229 0.814; 0.875; 0.204 Neutral Right vs. left PFC 18 0.605; 1.036; 0.238 0.876; 0.949; 0.229 0.858; 0.982; 0.367 0.814; 1.114; 0.268 Nurses Right PFC Affective vs. neutral 18 0.806; 0.430; 0.099 0.876; 1.565; 0.344 0.858; 1.544; 0.194 0.814; 1.516; 0.344 Left PFC Affective vs. neutral 18 0.605; 0.847; 0.194 0.876; 0.868; 0.194 0.858; 0.868; 0.115 0.814; 0.902; 0.212 Students Right PFC Affective vs. neutral 18 0.024; 3.298; 0.841* 0.907; 0.555; 0.153 0.928; 0.459; 0.176 0.948; 0.466; 0.098 Left PFC Affective vs. neutral 18 0.928; 0.125; 0.029 0.876; 0.793; 0.191 0.858; 0.809; 0.151 0.814; 0.997; 0.229 Affective Right PFC Nurses vs. students 36 0.928; − 0.092; 0.031 0.907; − 0.473; 0.158 0.928; − 0.440; 0.147 0.948; − 0.374; 0.125 Left PFC Nurses vs. students 36 0.605; 0.765; 0.255 0.995; 0.007; 0.002 0.982; 0.022; 0.007 0.961; 0.050; 0.017 Neutral Right PFC Nurses vs. students 36 0.400; 1.734; 0.578 0.876; − 0.798; 0.266 0.858; − 0.819; 0.273 0.814; − 0.840; 0.280 Left PFC Nurses vs. students 36 0.605; 0.845; 0.232 0.876; − 0.085; 0.028 0.982; − 0.054; 0.018 0.961; 0.132; 0.044 The values in table are displayed in (FDR-adjusted p-values; t-values; Cohen’s d). df represents degrees of freedom. * indicates FDR-adjusted p < 0.05. Weighted graph theory analysis Based on the two-way mixed ANOVA conducted separately on the clustering coefficient (CC), characteristic path length (λ), global efficiency (Eglobal), and local efficiency (Elocal), we found no significant (p > 0.05) results at the global level analysis. At the regional level, three-way mixed ANOVA conducted on the nodal efficiency (Enodal), CC, and λ revealed that none of the interaction effects (p > 0.05) were found to be significant among the three factors (group type, emotional state, and brain region). From the pairwise comparison, as depicted in Table 2, no significant result was observed (p > 0.05). Behavioral data From the two-way mixed ANOVA of the behavioral performance parameters, we found no significant results in the number of correctly solved questions, accuracy, and response time. The pairwise comparison results also did not demonstrate any significant differences between the two factors (emotional state and group type). Correlation between HRV and FC indices Based on the Pearson’s correlation analysis, the relationship between HRV and FC indices was assessed. From the results in Table 3, with the adjustment of FDR correction, we identified a significant negative moderate correlation (r = − 0.459, FDR-adjusted p = 0.020) between changes in the global SMP and changes in the RMSSD. However, neither of the weighted graph theory indices significantly correlated with the RMSSD score.Table 3 Results of correlation between all FC parameters and RMSSD. Parameters Correlation FDR-adjusted p-values SMP − 0.459 0.020* Eglobal − 0.284 0.098 Elocal − 0.278 0.098 CC − 0.277 0.098 λ − 0.275 0.098 *Represents FDR-adjusted p < 0.05. Discussion This study introduces a computation of FC (semi-metric analysis), which could be a more effective technique to assess fNIRS-based FC changes due to the affective and neutral emotional states. Firstly, group types were identified in terms of HRV and FC semi-metricity. The significant reduction in the RMSSD indicated a clear decrease in parasympathetic activity among the nursing students when in the affective state; on the other hand, the RMSSD of the registered nurses did not show any significant change despite changes in emotional states. In terms of behavioral performance, the nurses, exhibiting no significant changes in HRV and FC indices, had no significant difference in the number of correctly solved questions, accuracy, and response time. Likewise, the behavioral performance indices of the students were observed to be insignificant. Proceeding to the FC analyses, the comparison of semi-metric analysis and graph theory analysis in detecting emotional effects was evaluated based on two approaches, including ANOVA and correlation analysis with HRV. The conventional weighted graph theory analysis showed no significant results for both nursing students and registered nurse groups. On the other hand, by splitting into individual groups, the semi-metric analysis was able to distinguish significant changes of semi-metricity, especially at the right PFC among the students due to emotional effects. According to previous studies21,22, the significant reduction in HRV among nursing students might be explained by the adaptive physiological responses under the elicitation of external emotional stimuli. In contrast, the non-significant change in ANS activity among the registered nurses, as indicated by HRV values, might reflect that the nurses have developed their own coping strategy and the affective stimulus did not affect decision making (i.e., the task). This validated the induction of emotional states as a reference for FC analyses. Moving on to FC analyses, earlier fMRI studies determined that the increment in SMP reflected a higher level of hyperconnectivity and dispersal of FC, which included other brain regions10. The presence of hyperconnectivity in PFC areas has been further linked by previous studies to the processing of emotions such as anxiety and stress23,24. The apparent changes in brain semi-metricity in the right PFC regions among students might imply the increase in information sharing between the right PFC and other brain regions due to emotional effects. As discussed in several studies13,25, lateral PFC areas are involved in the cognitive control of emotion. Our ANOVA results showed an agreement with previous studies4,13,26 where the right lateralized asymmetry of FC was expected as the students were exposed to affective stimuli. Moreover, the study27 demonstrated that the non-significant emotional effect on task performance among students was due to the compensatory effort of the subjects. According to the attentional control theory28, students tended to maintain their behavioral performance in the affective state by increasing the executive function of PFC, which involved the right lateral PFC (BA 9/46), in agreement with our findings in the regional semi-metric analysis. This explained the emotional effects that were found to be significant in the change of brain topology, but not significant in behavioral performance among students. It also revealed that subjects in the affective state, indicated by a significant drop in the HRV RMSSD, possessed a significant reduction in the SMP. The linear correlation analysis strengthened the reliability of the semi-metric analysis by detecting a significant moderate negative association between the SMP and the RMSSD, as shown in Table 3. Therefore, it can be concluded that the affected cognition due to emotion may be detected from changes in the SMP. This study has some limitations. Firstly, the structural description of the semi-metric network remains unclear. In the present study, the semi-metric network was constructed based on the existence of indirect paths, but it is still a challenge to specify all involved paths along all indirect connections. Assessing the differences between the two emotional states may help identify the brain regions involved in emotional cognition. Secondly, the subject groups are significantly different (p < 0.001) in age. The age effect could possibly confound with the FC results of group comparisons. Thus, further work would include the structural study of the semi-metric network and the correction of age effect. Methods In this section, we present the overall functional connectivity analysis framework, which includes a novel FC analysis using semi-metric and benchmark FC analysis based on weighted graph theory, as illustrated in Fig. 3. The framework utilizes heart rate variability (HRV) to validate the success of inducing the affective state, and statistical tests such as ANOVA and correlation analysis with HRV analysis to compare the effectiveness of both the FC estimation methods.Figure 3 Summary flow chart for FC analyses and HRV analysis. Using the weighted network theory, the derived shortest path was directly quantified into network efficiency. In semi-metric metric analysis, the semi-metric component was extracted to be quantified as SMP. Proposed FC estimation method Data preprocessing First, the motion artifact was eliminated from the optical density (OD) of fNIRS signals using wavelet-based motion correction based on the hmrMotionCorrectWavelet function in HOMER229. The OD signal was decomposed into Gaussian distributed wavelet coefficients. Wavelet coefficients exceeding 1.5 times the interquartile range were eliminated as motion artifacts. By converting the corrected OD to ∆HbO and ∆HbR, we applied a low-pass filter with 1.0 Hz to remove high-frequency noise components. Subsequently, we extracted the functional neuronal component by separating the systemic physiological component (i.e., cerebral blood circulation) from the fNIRS signal based on the hemodynamic modality separation (HMS) method30. Functional connectivity matrix FC refers to the temporal correlation of the interacting cerebral region signals during the cognitive task31. In this study, we employed wavelet transform coherence (WTC) to construct brain FC matrices by using MATLAB Wavelet Coherence Toolbox32. WTC provided an advanced computation to Pearson’s correlation to measure the time-varying correlation between two signals in the frequency domain. It is suitable to assess non-stationary changes between fNIRS signals, especially the task-associated changes, and it has been widely used to investigate brain FC in fNIRS studies33,34. Based on the separation of the functional signal in the HMS method using the linear relationship assumption between ∆HbO and ∆HbR, we could expect the same FC matrices for both signal types. To verify our assumption in selecting signal types, we applied the same FC analyses and eventually observed the same results for both functional ∆HbO and ∆HbR signals. Therefore, we only selected functional ∆HbO signals, which are more sensitive to task-related changes, as the backbone of measurement35. The analyses of functional ∆HbR signals are shown in Supplementary Table 1 to Table 6. The functional ∆HbO time series signals were initially decomposed into wavelet coefficients in the time–frequency domain using wavelet transform36. Subsequently, we computed the pairwise correlation between all channels’ wavelet coefficients to construct time–frequency dimensional network correlation matrices. The frequency band of interest lies between 0.01 Hz and 0.2 Hz37. Within this range of frequency of interest, we extracted and averaged the 60 s task-relevant correlations to construct 32 × 32 network matrices. The channels represent the nodes, whereas the averaged correlation values denote weighted and undirected network edges. The constructed weighted network matrices were then submitted for semi-metric analysis and typical weighted graph theory analysis, as illustrated in Fig. 3. Semi-metric analysis From the weighted and undirected graphs, we converted the correlation matrix to a distance graph by using a distance conversion function38,39 as per Eq. (1): 1 lij=1xij-1 where li j denotes the distance from node i to j and xi j is the correlation weight between nodes i and j, given that the positions of the two different nodes are i to j. Subsequently, we labeled the semi-metric edges if li j was less than the summation of paths via other nodes between nodes i and j, for instance: lac < lab + lbc, given that there are nodes a, b, and c. As described in the pseudocode in Algorithm 1, we initiated the detection of semi-metric edges by finding the shortest paths, l′, based on the shortest path algorithm (i.e., Johnson’s Algorithm40). Next, we computed the ratio of semi-metricity, si j10: 2 sij=lijlij′ si j greater than 1 represented semi-metric edges whereas si j equivalent to 1 denoted metric backbone edges. Semi-metric properties in FC networks Finding the shortest paths in FC analysis may utilize two or more nodes to allow direct flow or sharing of information, respectively. Conventional graph theory, which quantifies an FC matrix based on its shortest paths, does not consider path sharing in the shortest paths. The application of semi-metric analysis categorized the shortest paths as either direct paths or sharing paths, as constructed in Fig. 4. A map of semi-metric edges shows information about sharing paths when the number of nodes involved, is greater than two. In detail, Fig. 5 shows that more than 20% of all the shortest paths were constructed by utilizing more than two-node paths, confirming a strong presence of path sharing in the shortest paths.Figure 4 FC maps showing the network matrices involved in conventional graph theory and semi-metric analysis. Further interpretation is conducted in semi-metric analysis where the information about sharing feature in a shortest path matrix is considered. Figure 5 The total number of shared paths used to construct the shortest paths among all subjects. Performance metric To characterize the semi-metric behavior of the brain network, we calculated the SMP based on the semi-metric ratio10: 3 SMP=∑i,jsij>1E where E is the total number of connections in the original network. Ultimately, SMP values were analyzed statistically at the global and regional levels based on the regions of interest (ROI). Weighted graph theory analysis Acting as a test bench for semi-metric analysis, we performed a typical weighted graph theory approach2,41 to explore the reliability of semi-metric analysis in differentiating emotional states. The weighted graph theory analysis was conducted using the Brain Connectivity Toolbox (BCT)5. From the unthresholded and weighted functional network, we performed a global analysis by computing the network topological parameters including the CC, λ, Eglobal, and Elocal. When the other nodes around a node of interest form at least a triangular connection, the measure of the cliquishness is defined as CC42 in Eq. (4): 4 CC=1N∑i2mki(ki-1) From the FC matrices, we computed λ to quantify the integration of the potential information flow based on the average shortest path length as per Eq. (5): 5 λ=1N∑i∈G∑j∈G,j≠idijN-1 Eglobal was described as the inverse of the harmonic mean of the shortest path length within the whole network43,44. It quantified the ability of global and concurrent exchange of information between connected edges as derived in Eq. (6): 6 Eglobal=1N(N-1)∑j≠i∈G1dij The Elocal of a network G is defined as the mean local efficiency of each node41,43 as shown in Eq. (7) It not only characterizes the capability of information flow across node i to its nearest neighbor nodes but also reflects the tolerance of neighboring nodes when there is a defect in node i. 7 Elocal=1N∑i∈GEglobal(Gi) We further decomposed the PFC networks into regional subgraphs based on the ROI. To quantify the information propagation ability across regions, we evaluated the regional CC, λ, and Enodal of all nodes within the ROI using Eq. (8)43: 8 Enodal=1N-1∑j∈G1dij From the equations above, i = 1, 2, 3, N; j ≠ i refers to the region relative to node i; m is the number of neighboring edges; di j denotes the weighted shortest path length between nodes i and j; N refers to the total number of nodes in the network, G, which consists of all the nodes. Validation experiment Subjects In total, 39 right-handed, healthy nursing subjects, consisting of 19 nurses with actual working experience (Edinburgh Handedness Inventory45 scale = 86.18 ± 15.53, age = 30.44 ± 3.20 years old, working experience = 8.32 ± 3.04 years) and 20 students with only internship experience (Edinburgh Handedness Inventory scale = 93.13 ± 12.48, age = 20.68 ± 0.82 years old, internship experience = 2.70 ± 0.41 years) participated in this study. Prior to the experiment, all subjects had to complete a screening questionnaire which included demographic information such as physical health, mental condition and family history of disease. Subjects with known history of any psychiatric or neurological disorders were excluded. The participants were prohibited from consuming alcohol and caffeine, smoking, and exercising for at least 3 h before the experiment. One nursing student who did not fulfill the requirements was excluded. Using G*Power 346, a sensitivity power analysis was carried out to evaluate the sample size based on the repeated measure ANOVA (within-between interaction), given the following conditions: (1) significant level = 0.05, (2) power of 1 − β = 0.80, (3) 2 groups, and (4) 2 measurements. The generated minimal detectable effect reported a critical effect size f(U) = 0.480 (or ηp2 =  0.102). This study was approved by the ethics committee of Universiti Kuala Lumpur Royal College of Medicine Perak (UniKL RCMP) (approval number: UniKLRCMP/MREC/2018/018). All the subjects provided informed consent, and the experiment was carried out in accordance with the Declaration of Helsinki guidelines and regulations. Measurement Brain activity in the PFC was measured using a dual-wavelength (695 nm and 830 nm) multichannel OT-R40 fNIRS continuous wave system (Hitachi Medical Corporation, Japan), with a sampling rate of 10 Hz. A 52-channel 3 × 11 optodes layout (17 sources and 16 detectors) with a source-detector distance of 3 cm was deployed based on the international 10/20 system47 along the T4-Fpz-T3 positions. By using distinctive absorption coefficients of different chromophores and the modified Beer-Lambert Law48, we calculated the change in the concentration of oxygenated hemoglobin (∆HbO) and deoxygenated hemoglobin (∆HbR) based on the changes in light intensity of the dual-wavelength light. We estimated the channel localization according to the Montreal Neurological Institute (MNI) coordination, determining the Brodmann area (BA) for each channel35,47. Here, we identified PFC regions based on the 32 channels as labeled in Fig. 6 and subsequently divided the regions into the left and right PFC as our ROI. Measurements from channels 16 and 37 were excluded when we compared the two hemispheres.Figure 6 Probes were setup on subjects’ forehead and scalp based on the international 10/20 system. A Nellcor DS-100A ear clip sensor was placed on the left ear of the subjects. The earclip sensor was connected to the AFE4490SPO2EVM Evaluation Board (Texas Instruments Inc., Dallas, Texas) to collect photoplethysmographic (PPG) signals at a sampling rate of 200 Hz simultaneously with the fNIRS measurement. The purpose of measuring PPG signals was to perform heart rate variability (HRV) analysis. Compared with electrocardiogram (ECG), PPG offers higher simplicity, minimum subject discomfort, and lower cost. Previous studies have reached a consensus that PPG is an alternative to ECG in estimating HRV49,50. The evaluation of emotional states based on PPG has also been implemented in a recent study51. Affective and neutral emotional stimuli Two sets of the auditory emotional stimuli were retrieved to induce different emotional states from the International Affective Digitized Sounds system (IADS)52. The first set consisting of ten affective sound clips was referred to as the "case" set while the other set of ten neutral sound clips was labelled as the "control" set. The classes of stimuli were based on the emotional circumplex model, as shown in Fig. 753. Descriptively, the emotional circumplex model comprises two independent neurophysiological dimensions, known as valence and arousal ratings. These ratings are scaled according to the Self-Assessment Manikin (SAM) 9-point ratings54. The IADS provides a standardized database of emotional stimuli based on two-dimensional ratings. We defined affective stimulus as sound clips audible in hospital with negative valence (rating of 2.147 ± 0.473 out of 9) and high arousal (rating of 7.388 ± 0.494 out of 9) including an ambulance siren, the crying of a baby, and human screams, whereas the neutral stimulus comprised sound clips with neutral valence (rating of 5.197 ± 0.720 out of 9) and medium arousal (rating of 4.560 ± 0.380 out of 9), such as the sounds of typewriting, clock ticking, and raindrops. The stimulus was played in the background throughout the entire task to induce different emotional states (one session, one emotional state).Figure 7 Emotional circumplex model based on the 9-point Self-Assessment Manikin (SAM) ratings classifies various emotional adjectives. The red-shaded and blue-shaded regions indicate the regions of retrieving affective and neutral stimuli respectively. Task This experiment consisted of two sessions differentiated by two sets of auditory emotional stimuli. All subjects performed the second experimental session at least six weeks after the first session. They repeated the experiment with another auditory emotional stimulus set. The order of the sessions was counterbalanced across the subjects. In each session, subjects sat approximately 60 cm in front of a monitor in a quiet, dimly lit room. As shown in Fig. 8, the experiment started with 20 s of rest followed by five alternate periods of task and rest. The subjects were required to focus on the on-screen cross and relax. During each 60 s task period, up to five questions about the nursing case study with four choices were displayed in succession on the monitor. Subjects were instructed to answer swiftly and complete as many questions as possible within a task period of 60 s. At the same time, auditory emotional stimuli were played through a speaker during the task periods. The types of the questions were retrieved based on five objectives proposed in Bloom’s taxonomy, including remembering, understanding, applying, analyzing, and evaluating55. The questions asked in both sessions were standardized in terms of type and difficulty level. The number of correctly solved questions, accuracy (percentage of correct answers over the total number of attempts), and response time were recorded and included in the statistical analysis.Figure 8 Task paradigm constructed for validation experiment. HRV analysis The HRV is described as the fluctuation of distance between two successive heart beats (also known as normal-to-normal (NN) interval)21,56. HRV has been widely used as a quantitative marker to investigate the human autonomic nervous system (ANS) responses. Functioning as a physiological indicator of emotion processing57,58, it provides a non-invasive means to determine the balance between sympathetic (fight or flight) and parasympathetic (rest and digest) activity. The standard HRV analysis can be derived in both time and frequency domains21,59. In this study, we focused on the time-domain HRV analysis as it demonstrated a better accuracy for short-term HRV recording60. We computed the RMSSD between normal heartbeats. Low RMSSD has been found to be associated with low parasympathetic activity due to poor emotional regulation56,59. We conducted HRV analysis using MATLAB-based (MathWorks Inc., Natick, MA) HRVTool v1.04 (https://github.com/MarcusVollmer/HRV)61. Firstly, as depicted in Fig. 3, the PPG signals were smoothened using moving average filter with window length equivalent to the sampling rate i.e. 200 data points. By setting the maximum and minimum heart rate to 180 and 60 beats per minute, respectively, the NN intervals were extracted from the preprocessed signals by using the QRS detection algorithm62. The artifacts (abnormal NN interval) were then eliminated using the same filtering method applied by Vollmer61. Ultimately, from the filtered NN intervals, we computed the HRV RMSSD by using the following formula: 9 RMSSD=1n-1∑i=1n-1(NNi+1-NNi)2 where NNi denotes the time intervals of successive beats and n denotes the total number of normal peaks. Statistical analysis Statistical analysis was conducted on the subjects’ behavioral performance, HRV and FC data. All multiple comparisons were FDR-adjusted using Benjamini and Hochberg method63 at desired q-level (FDR-adjusted p = 0.05). Behavioral performance We applied two-way mixed ANOVA to evaluate the group and emotional state effects on the behavioral performance indices (the number of correctly solved questions, accuracy and response time). HRV To examine the statistical differences in the RMSSD, we performed two-way mixed ANOVA to evaluate the interaction between emotional state and group type. Comparison of functional connectivity methods We compared the proposed semi-metric analysis to the weighted graph theory analysis in discriminating the emotional effect based on (1) ANOVA (2) correlation analysis. In the entire PFC analysis, using IBM SPSS Statistics v23 (IBM Corp, Armonk, NY), two-way mixed ANOVA was conducted on the FC indices (SMP, Eglobal, Elocal, CC, and λ) to examine the group and emotional task as the between-, and within-subjects factor respectively. At the regional level of semi-metric analysis and weighted graph theory analysis, three-way mixed ANOVA was carried out to evaluate the interaction between factors of group type, emotional state, and brain region (asymmetry). The effect size was determined using the partial eta squared (ηp2) and Cohen’s d for ANOVAs and pairwise comparisons, respectively. Furthermore, to examine the association between stimulated emotional states and FC indices, we computed the Pearson’s correlation, r between the average changes in the RMSSD and the average changes in the global SMP, Eglobal, Elocal, CC, and λ in affective versus neutral emotional states. Conclusion This study explored semi-metric in analyzing fNIRS-based functional connectivity. The semi-metric analysis characterized the weighted FC by considering the information sharing paths at the global and regional levels of FC. The experimental results revealed that the semi-metric analysis, as correlated to HRV, was able to detect that the nursing students were more susceptible to emotional change. Under the affective condition, the nursing students demonstrated significant change in semi-metricity, but not in the conventional graph theory analysis. The results suggest the semi-metric analysis as an FC analytical technique could be generalized and extended to fNIRS. Further investigation on the age effect will help better understand about the underlying causes of reduced emotional sensitivity among the registered nurses. Supplementary Information Supplementary Information. Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary Information The online version contains supplementary material available at 10.1038/s41598-020-79053-z. Acknowledgements The project was funded by Ministry of Higher Education Malaysia under HICoE scheme to CISIR, UTP. Author contributions T.B.T., E.E., and Y.L.C. designed the study. J.S.C., Y.L.C., and H.Y.C. acquired the data. J.S.C., T.B.T., M.K., and C.L. analyzed the data. J.S.C. and T.B.T. wrote the article. T.B.T. supervised the manuscript preparation, revision and submission. All authors have reviewed the manuscript. Competing interests The authors declare no competing interests. ==== Refs References 1. Bullmore E Sporns O Complex brain networks: Graph theoretical analysis of structural and functional systems Nat. Rev. Neurosci. 2009 10 186 198 10.1038/nrn2575 19190637 2. Geng S Liu X Biswal BB Niu H Effect of resting-state fNIRS scanning duration on functional brain connectivity and graph theory metrics of brain network Front. Neurosci. 2017 11 392 10.3389/fnins.2017.00392 28775676 3. Vecchio F Miraglia F Maria Rossini P Connectome: Graph theory application in functional brain network architecture Clin. Neurophysiol. Pract. 2017 2 206 213 10.1016/j.cnp.2017.09.003 30214997 4. Sinha R Lacadie CM Constable RT Seo D Dynamic neural activity during stress signals resilient coping Proc. Natl. Acad. Sci. 2016 113 8837 8842 10.1073/pnas.1600965113 27432990 5. Rubinov M Sporns O Complex network measures of brain connectivity: Uses and interpretations Neuroimage 2010 52 1059 1069 10.1016/j.neuroimage.2009.10.003 19819337 6. Meunier D Lambiotte R Bullmore ET Modular and hierarchically modular organization of brain networks Front. Neurosci. 2010 4 200 10.3389/fnins.2010.00200 21151783 7. Kalavri V Simas T Logothetis D The shortest path is not always a straight line Proc. VLDB Endow. 2016 9 672 683 10.14778/2947618.2947623 8. Peeters S Semi-metric analysis of the functional brain network: Relationship with familial risk for psychotic disorder NeuroImage Clin. 2015 9 607 616 10.1016/j.nicl.2015.10.003 26740914 9. Simas, T. & Rocha, L. M. Semi-metric networks for recommender systems. In IEEE/WIC/ACM International Conferences on Web Intelligence and Intelligent Agent Technology, Vol. 3, 175–179 (2012). 10. Simas T Semi-metric topology of the human connectome: Sensitivity and specificity to autism and major depressive disorder PLoS ONE 2015 10 1 20 10.1371/journal.pone.0136388 11. Suckling J A winding road: Alzheimer’s disease increases circuitous functional connectivity pathways Front. Comput. Neurosci. 2015 9 140 10.3389/fncom.2015.00140 26635593 12. De La Iglesia-vaya, M., Molina-Mateo, J., Escarti-Fabra, M. J., Kanaan, A. S. & Martí-Bonmatí, L. Brain connections–resting state fmri functional connectivity. In Novel Frontiers of Advanced Neuroimaging (ed. Fountas, K. N.) 236–286 (IntechOpen, 2013). 13. Uchida M Emotion regulation ability varies in relation to intrinsic functional brain architecture Soc. Cogn. Affect. Neurosci. 2014 10 1738 1748 10.1093/scan/nsv059 14. Cui X Bray S Bryant DM Glover GH Reiss AL A quantitative comparison of NIRS and fMRI across multiple cognitive tasks Neuroimage 2011 54 2808 2821 10.1016/j.neuroimage.2010.10.069 21047559 15. Scarapicchia V Brown C Mayo C Gawryluk JR Functional magnetic resonance imaging and functional near-infrared spectroscopy: Insights from combined recording studies Front. Hum. Neurosci. 2017 11 419 10.3389/fnhum.2017.00419 28867998 16. Einalou Z Maghooli K Setarehdan SK Akin A Graph theoretical approach to functional connectivity in prefrontal cortex via fnirs Neurophotonics 2017 4 1 10.1117/1.NPh.4.4.041407 17. Mizuno M Hiroyasu T Hiwa S A functional nirs study of brain functional networks induced by social time coordination Brain Sci. 2019 9 1 11 10.3390/brainsci9020043 18. Di Martino V RelationshipBetweenWorkStressandWorkplaceViolenceintheHealthSector 2003 Geneva ILO 19. Bakker EJ Physical and mental determinants of dropout and retention among nursing students: Protocol of the spring cohort study BMC Nurs. 2018 17 27 10.1186/s12912-018-0296-9 29977155 20. Matthews, G., Wohleber, R. & Lin, J. Stress, skilled performance, and expertise: Overload and beyond. In The Oxford Handbook of Expertise (eds. Ward, P., Schraagen, J. M., Gore, J. & Roth, E.) 1–39 (Oxford University Press, 2019). 21. Kim HG Cheon EJ Bai DS Lee YH Koo BH Stress and heart rate variability: A meta-analysis and review of the literature Psychiatry Investig. 2018 15 235 245 10.30773/pi.2017.08.17 29486547 22. Hammoud S Karam R Mourad R Saad I Kurdi M Stress and heart rate variability during university final examination among lebanese students Behav. Sci. 2019 9 3 10.3390/bs9010003 23. Bartholomew ME Yee CM Heller W Miller GA Spielberg JM Reconfiguration of brain networks supporting inhibition of emotional challenge Neuroimage 2019 186 350 357 10.1016/j.neuroimage.2018.10.066 30394327 24. Dunkley BT Wong SM Jetly R Wong JK Taylor MJ Post-traumatic stress disorder and chronic hyperconnectivity in emotional processing NeuroImage Clin. 2018 20 197 204 10.1016/j.nicl.2018.07.007 30094169 25. Glotzbach E Prefrontal brain activation during emotional processing: A functional near infrared spectroscopy study (fNIRS) Open Neuroimaging J. 2011 5 33 39 10.2174/1874440001105010033 26. Davidson R Fox N Asymmetrical brain activity discriminates between positive and negative affective stimuli in human infants Science 1982 218 1235 1237 10.1126/science.7146906 7146906 27. Ozawa S Matsuda G Hiraki K Negative emotion modulates prefrontal cortex activity during a working memory task: A NIRS study Front. Hum. Neurosci. 2014 8 1 10 10.3389/fnhum.2014.00046 24474914 28. Eysenck MW Derakshan N Santos R Calvo MG Anxiety and cognitive performance: Attentional control theory Emotion 2007 7 336 353 10.1037/1528-3542.7.2.336 17516812 29. Molavi B Dumont GA Wavelet-based motion artifact removal for functional near-infrared spectroscopy Physiol. Meas. 2012 33 259 270 10.1088/0967-3334/33/2/259 22273765 30. Yamada T Umeyama S Matsuda K Separation of fnirs signals into functional and systemic componentsbased on differences in hemodynamic modalities PLoS ONE 2012 7 e50271 10.1371/journal.pone.0050271 23185590 31. Varela F Lachaux J-P Rodriguez E Martinerie J The brain web: Phase large-scale integration Nat. Rev. Neurosci. 2001 2 229 239 10.1038/35067550 11283746 32. Grinsted A Moore JC Jevrejeva S Application of the cross wavelet transform and wavelet coherence to geophysical time series Nonlinear Process. Geophys. 2004 11 561 566 10.5194/npg-11-561-2004 33. Cui X Bryant DM Reiss AL NIRS-based hyperscanning reveals increased interpersonal coherence in superior frontal cortex during cooperation Neuroimage 2012 59 2430 2437 10.1016/j.neuroimage.2011.09.003 21933717 34. Baker JM Bruno JL Gundran A Hosseini SMH Reiss L fNIRS measurement of cortical activation and functional connectivity during a visuospatial working memory task PLoS ONE 2018 13 e0201486 10.1371/journal.pone.0201486 30071072 35. Sato H A nirs-fmri investigation of prefrontal cortex activity during a working memory task Neuroimage 2013 83 158 173 10.1016/j.neuroimage.2013.06.043 23792984 36. Zhang, X. et al. Activation detection in fnirs by wavelet coherence. In Proc. SPIE 8317, Med. Imaging 2012: Biomed. Appl. Mol. Struct. Funct. Imaging, vol. 8317, 1–7 (2012). 37. Pinti P Scholkmann F Hamilton A Burgess P Tachtsidis I Current status and issues regarding pre-processing of fNIRS neuroimaging data: An investigation of diverse signal filtering methods within a general linear model framework Front. Hum. Neurosci. 2019 12 1 21 10.3389/fnhum.2018.00505 38. Simas T Rocha LM Distance closures on complex networks Net. Sci. 2015 3 227 268 10.1017/nws.2015.11 39. Dombi J A general class of fuzzy operators, the demorgan class of fuzzy operators and fuzziness measures induced by fuzzy operators Fuzzy Sets Syst. 1982 8 149 163 10.1016/0165-0114(82)90005-7 40. Johnson DB Efficient algorithms for shortest paths in sparse networks J. ACM 1977 24 1 13 10.1145/321992.321993 41. Latora V Marchiori M Economic small-world behavior in weighted networks Eur. Phys. J. B 2003 32 249 263 10.1140/epjb/e2003-00095-5 42. Pedersen M Omidvarnia AH Walz JM Jackson GD Increased segregation of brain networks in focal epilepsy: An fMRI graph theory finding NeuroImage Clin. 2015 8 536 542 10.1016/j.nicl.2015.05.009 26110111 43. Lo CY Diffusion tensor tractography reveals abnormal topological organization in structural cortical networks in Alzheimer’s disease J. Neurosci. 2010 30 16876 16885 10.1523/JNEUROSCI.4136-10.2010 21159959 44. Latora V Marchiori M Efficient behavior of small-world networks Phys. Rev. Lett. 2001 87 198701-1 198701-4 10.1103/PhysRevLett.87.198701 11690461 45. Oldfield RC The assessment and analysis of handedness: The Edinburgh inventory Neuropsychologia 1971 9 97 113 10.1016/0028-3932(71)90067-4 5146491 46. Faul F Erdfelder E Lang A-G Buchner A G* power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences Behav. Res. Methods 2007 39 175 191 10.3758/BF03193146 17695343 47. Singh AK Okamoto M Dan H Jurcak V Dan I Spatial registration of multichannel multi-subject fNIRS data to mni space without MRI Neuroimage 2005 27 842 851 10.1016/j.neuroimage.2005.05.019 15979346 48. Kocsis L Herman P Eke A The modified Beer–Lambert law revisited Hum. Physiol. 2006 51 91 98 49. Selvaraj N Jaryal A Santhosh J Deepak KK Anand S Assessment of heart rate variability derived from finger-tip photoplethysmography as compared to electrocardiography J. Med. Eng. Technol. 2008 32 479 484 10.1080/03091900701781317 18663635 50. Lu S Can photoplethysmography variability serve as an alternative approach to obtain heart rate variability information? J. Clin. Monit. Comput. 2008 22 23 29 10.1007/s10877-007-9103-y 17987395 51. Choi K-H Is heart rate variability (HRV) an adequate tool for evaluating human emotions?—A focus on the use of the international affective picture system (IAPS) Psychiatry Res. 2017 251 192 196 10.1016/j.psychres.2017.02.025 28213189 52. Bradley, M. M. & Lang, P. J. The international affective digitized sounds (2nd edition; iads-2): affective ratings of sounds and instruction manual. Univ. Florida, Gainesville, Fl. Technical report B-3. (2007). 53. Posner J Russell JA Peterson BS The circumplex model of affect: An integrative approach to affective neuroscience, cognitive development, and psychopathology Dev. Psychopathol. 2005 17 715 734 10.1017/S0954579405050340 16262989 54. Bradley MM Lang PJ Measuring emotion: The self-assessment manikin and the semantic differential J. Behav. Ther. Exp. Psychiatry 1994 25 49 59 10.1016/0005-7916(94)90063-9 7962581 55. Adams NE Bloom’s taxonomy of cognitive learning objectives J. Med. Libr. Assoc. 2015 103 152 153 10.3163/1536-5050.103.3.010 26213509 56. Michels N Children’s heart rate variability as stress indicator: Association with reported stress and cortisol Biol. Psychol. 2013 94 433 440 10.1016/j.biopsycho.2013.08.005 24007813 57. Nardelli M Valenza G Greco A Lanata A Scilingo EP Recognizing emotions induced by affective sounds through heart rate variability IEEE Trans. Affect. Comput. 2015 6 385 394 10.1109/TAFFC.2015.2432810 58. Brouwer AM van Wouwe N Mühl C van Erp J Toet A Perceiving blocks of emotional pictures and sounds: Effects on physiological variables Front. Hum. Neurosci. 2013 7 1 10 23355817 59. Shi H Differences of heart rate variability between happiness and sadness emotion states: A pilot study J. Med. Biol. Eng. 2017 37 527 539 10.1007/s40846-017-0238-0 60. Nunan D Sandercock GR Brodie DA A quantitative systematic review of normal values for short-term heart rate variability in healthy adults PACE Pacing Clin. Electrophysiol. 2010 33 1407 1417 10.1111/j.1540-8159.2010.02841.x 20663071 61. Vollmer M A robust, simple and reliable measure of heart rate variability using relative RR intervals Comput. Cardiol. 2015 42 609 612 62. Pan J Tompkins WJ A real-time QRS detection algorithm IEEE Trans. Biomed. Eng. 1985 32 230 236 10.1109/TBME.1985.325532 3997178 63. Benjamini Y Hochberg Y Controlling the false discovery rate: A practical and powerful approach to multiple testing J. R. Stat. Soc. Ser. B 1995 57 289 300