# A Machine Learning Approach Involving Functional Connectivity Features to Classify Rest-EEG Psychogenic Non-Epileptic Seizures from Healthy Controls

^{1}

^{2}

^{3}

^{4}

^{5}

^{6}

^{7}

^{8}

^{*}

## Abstract

**:**

## 1. Introduction

- Investigating possible intrinsic alterations in resting-state brain networks oscillations of patients with PNES using PSD analysis.
- Determining whether functional connectivity alterations of PNES subjects could be associated with specific areas which lead to regional network dysfunctions in local oscillations, as well as inter-regional synchronization.
- Investigating a machine learning approach, involving rest EEG-based functional connectivity features, to disentangle PNES from non-PNES subjects.

## 2. Related Works

## 3. Materials and Methods

#### 3.1. Participants

#### 3.2. EEG Recording

#### Preprocessing

#### 3.3. Power Spectral Density Analysis

#### 3.4. Graph Analysis

#### 3.5. Phase-Locking Index Analysis

#### 3.5.1. Network Parameters

#### 3.6. Graph Metrics

#### 3.6.1. Averaged Shortest Path Length

#### 3.6.2. Global Efficiency

#### 3.6.3. Clustering Coefficient

#### 3.6.4. Small-Worldness

#### 3.6.5. Node Betweenness

**i**and

**j**, and ${\lambda}_{i,j}\left(i\right)$ is the number of shortest paths making use of the node

**i**.

#### 3.7. Statistical Analysis

## 4. Classification

#### 4.1. Dataset Preparation

#### 4.2. Proposed Machine Learning Classifier

- SVM classifier: SVM technique is a computer algorithm that learns, based on a statistical theories, labels assigned to objects [49,50]. SVM technique attempts to find a hyperplane that provides the best separation between classes of points. In this study, a SVM classifier with a linear kernel is implemented. The mathematical background on SVM is reported in detail in Reference [51]. SVM classifier is suitable to work with 2D datasets; thus, it was fed with an R (raw) ∗ C (column) matrix set. Here, (R = 39) and (C = 4.801). However, after training/test dataset selection, here, 70% for training and 30% for testing, it was submitted for 8 k-fold cross-validations. The SVM classifier was implemented in python using scikit-learn packages. The parameter C is a hypermeter in SVM used to control the error of class separation. Here, we used C = 0.01.
- LDA classifier: Linear Discriminant Analysis uses a statistical methods applied for data classification and dimensionality reduction. LDA reduces the data dimensionality in order to improve the class separability. LDA projects clusters of data into lower dimensional space to increase class separability by decreasing intraclass differences. More mathematical detail on LDA background is reported in Reference [52]. Our LDA classifier is suitable to work with 2D datasets; thus, it was fed with an R (raw) ∗ C (column) matrix set. Here, (R = 39) and (C = 4.801). The LDA classifier was implemented in python using scikit-learn packages.
- MLP classifier: Multilayer Perceptron is a supervised feed-forward neural network commonly used for classification and regression tasks [53,54]. We designed an MLP classifier with two hidden layers with 18 and 4 neurons, respectively. The first hidden layer was designed with ReLU function, whereas the latter hidden layer implements a softmax for binary classification. The training procedure was submitted to k-fold cross validation. MLP was implemented in python using scikit-learn packages. Our MLP architecture is depicted in detail in Figure 2. In this paper, the MLP is trained using supervised learning mode for ${10}^{3}$ epochs on a MacBook Pro 2.2 GHz Intel Core i7 quad-core (training time ≈ 480 s). The features vector (sized 1 × 39) is used as input to a MLP with 2 hidden layers of 39 and 18 hidden units, respectively. The ReLU is used as activation function for each hidden neuron. The network ends with a softmax output layer to perform a binary classification task: PNES versus HC. The architecture here was referred as ${\mathrm{MLP}}_{(39,18)}$.

## 5. Results

#### 5.1. Relative PSD Analysis

^{2}/Hz] log normalize values. Furthermore, in beta band, we founded PSD values between [0.025, 0.25 μV

^{2}/Hz], indicating that most energies focused at lower frequencies when the subjects were resting. Comparing the relative PSD values between the PNES group and the HC (see Figure 3A), we found higher PSD in the delta and theta band for PNES more than HC. Conversely, in the alpha band (see Figure 3B), the PSD results are higher for HC. To infer more insight into the power spectrum results, we performed a post–hoc analysis in four frequency bands and three brain parcellations (Frontal, Central, and Parieto-occipital areas) (see Figure 4). In this paper, we selected (Fp1-Fp2-F3-F4) sensors for frontal area, (C3-C4-Cz-Pz) sensors for the central area, and the sensors (P3-P4-T3-T4-T5-T6-O1-O2-P3-P4) for parieto-occipital. The PNES group was found to have significantly higher PSD in the delta band in the frontal and central area (p < 0.05), and in the delta and theta band. Additionally, we found an increased PSD in alpha and beta for HC, more than PNES in the frontal, central, and parieto-occipital areas (p < 0.01). In the theta band, post-hoc analysis showed that the PNES group differed as compared to HC (p = 0.0082, p = 0.0476, and p = 0.0076, respectively, for the frontal, central, and parieto-occipital area). Differences were also present in the beta band (p = 0.0255 and p = 0.0308 and p = 0.0043, respectively). In the beta band, we found higher values of PSD for HC compared to PNES. In the delta band, HC showed a lower PSD values as compared to HC (p = 0.0262, p = 0.342, p = 0.0153). In the alpha band, the values of PSD for HC were higher than PNES with (p = 0.0032, p = 0.00443, and p = 0.023, respectively) in all the brain parcellations. Compared to HC, the relative PSD values were increased for PNES in the delta band in the frontal and central areas. In contrast, PSD for alpha and beta bands were increased in the HC more than the PNES subjects in all the parcellations. Compared to HC, the relative PSD values were increased for PNES in the delta band in the frontal and central areas. As illustrated in Figure 4, frontal, the central, and parieto-occipital area displayed a significant PSD difference in the delta and theta bands.

#### 5.2. PLI Analysis

#### 5.2.1. Measures of Integration and Segregation

#### 5.2.2. Measures of Centrality

#### 5.3. Binary Classification

## 6. Discussion

#### 6.1. PSD Measures

#### 6.2. PLI Measures

#### 6.3. Measures of Segregation and Integration

#### 6.4. Measures of Centrality

#### 6.5. Small-Worldness

#### 6.6. Classification

## 7. Conclusions

## Author Contributions

## Funding

## Acknowledgments

## Conflicts of Interest

## References

- Stewart, R.S.; Lovitt, R.; Stewart, M. Are hysterical seizures more than hysteria? a research diagnostic criteria, DSM-III, and psychometric analysis. Am. J. Psychiatry
**1982**, 139, 926–929. [Google Scholar] [PubMed] - Vanderzant, C.; Giordani, B.; Berent, S.; Dreifuss, F.; Sackellares, J. Personality of patients with pseudoseizures. Neurology
**1986**, 36, 664. [Google Scholar] [CrossRef] [PubMed] - Alessi, R.; Valente, K.D. Psychogenic non-epileptic seizures at a tertiary care center in Brazil. Epilepsy Behav.
**2013**, 26, 91–95. [Google Scholar] [CrossRef] [Green Version] - Hesdorffer, D.C. Comorbidity between neurological illness and psychiatric disorders. CNS Spectrums
**2016**, 21, 230–238. [Google Scholar] [CrossRef] - Duncan, R.; Razvi, S.; Mulhern, S. Newly presenting psychogenic nonepileptic seizures: Incidence, population characteristics, and early outcome from a prospective audit of a first seizure clinic. Epilepsy Behav.
**2011**, 20, 308–311. [Google Scholar] [CrossRef] - Benbadis, S.R.; Hauser, W.A. An estimate of the prevalence of psychogenic non-epileptic seizures. Seizure
**2000**, 9, 280–281. [Google Scholar] [CrossRef] [Green Version] - LaFrance, W.C., Jr.; Baker, G.A.; Duncan, R.; Goldstein, L.H.; Reuber, M. Minimum requirements for the diagnosis of psychogenic nonepileptic seizures: A staged approach: A report from the International League Against Epilepsy Nonepileptic Seizures Task Force. Epilepsia
**2013**, 54, 2005–2018. [Google Scholar] [CrossRef] [PubMed] - Goldstein, L.H.; Mellers, J.D. Recent developments in our understanding of the semiology and treatment of psychogenic nonepileptic seizures. Curr. Neurol. Neurosci. Rep.
**2012**, 12, 436–444. [Google Scholar] [CrossRef] [PubMed] - Horton, J.C.; Adams, D.L. The cortical column: A structure without a function. Philos. Trans. R. Soc. B Biol. Sci.
**2005**, 360, 837–862. [Google Scholar] [CrossRef] [PubMed] - Barzegaran, E.; Joudaki, A.; Jalili, M.; Rossetti, A.O.; Frackowiak, R.S.; Knyazeva, M.G. Properties of functional brain networks correlate with frequency of psychogenic non-epileptic seizures. Front. Hum. Neurosci.
**2012**, 6, 335. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Xue, Q.; Wang, Z.Y.; Xiong, X.C.; Tian, C.Y.; Wang, Y.P.; Xu, P. Altered brain connectivity in patients with psychogenic non-epileptic seizures: A scalp electroencephalography study. J. Int. Med. Res.
**2013**, 41, 1682–1690. [Google Scholar] [CrossRef] - Li, R.; Li, Y.; An, D.; Gong, Q.; Zhou, D.; Chen, H. Altered regional activity and inter-regional functional connectivity in psychogenic non-epileptic seizures. Sci. Rep.
**2015**, 5, 11635. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Varone, G.; Gasparini, S.; Ferlazzo, E.; Ascoli, M.; Tripodi, G.G.; Zucco, C.; Calabrese, B.; Cannataro, M.; Aguglia, U. A Comprehensive Machine-Learning-Based Software Pipeline to Classify EEG Signals: A Case Study on PNES vs. Control Subjects. Sensors
**2020**, 20, 1235. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Zucco, C.; Calabrese, B.; Sturniolo, M.; Gambardella, A.; Cannataro, M. A Software Pipeline for Pre-Processing and Mining EEG Signals: Application in Neurology. In Proceedings of the SEBD 2021: The 29th Italian Symposium on Advanced Database Systems, Pizzo Calabro, Italy, 9 September 2021. [Google Scholar]
- Knyazeva, M.G.; Jalili, M.; Frackowiak, R.S.; Rossetti, A.O. Psychogenic seizures and frontal disconnection: EEG synchronisation study. J. Neurol. Neurosurg. Psychiatry
**2011**, 82, 505–511. [Google Scholar] [CrossRef] - Barzegaran, E.; Carmeli, C.; Rossetti, A.O.; Frackowiak, R.S.; Knyazeva, M.G. Weakened functional connectivity in patients with psychogenic non-epileptic seizures (PNES) converges on basal ganglia. J. Neurol. Neurosurg. Psychiatry
**2016**, 87, 332–337. [Google Scholar] [CrossRef] - Umesh, S.; Tikka, S.K.; Goyal, N.; Sinha, V.K.; Nizamie, S.H. Aberrant gamma band cortical sources and functional connectivity in adolescents with psychogenic non-epileptic seizures: A preliminary report. Psychiatry Res.
**2017**, 247, 51–54. [Google Scholar] [CrossRef] [PubMed] - Van der Kruijs, S.J.; Bodde, N.M.; Vaessen, M.J.; Lazeron, R.H.; Vonck, K.; Boon, P.; Hofman, P.A.; Backes, W.H.; Aldenkamp, A.P.; Jansen, J.F. Functional connectivity of dissociation in patients with psychogenic non-epileptic seizures. J. Neurol. Neurosurg. Psychiatry
**2012**, 83, 239–247. [Google Scholar] [CrossRef] [Green Version] - Ding, J.R.; An, D.; Liao, W.; Li, J.; Wu, G.R.; Xu, Q.; Long, Z.; Gong, Q.; Zhou, D.; Sporns, O.; et al. Altered functional and structural connectivity networks in psychogenic non-epileptic seizures. PLoS ONE
**2013**, 8, e63850. [Google Scholar] [CrossRef] - Van der Kruijs, S.J.; Jagannathan, S.R.; Bodde, N.M.; Besseling, R.M.; Lazeron, R.H.; Vonck, K.E.; Boon, P.A.; Cluitmans, P.J.; Hofman, P.A.; Backes, W.H.; et al. Resting-state networks and dissociation in psychogenic non-epileptic seizures. J. Psychiatr. Res.
**2014**, 54, 126–133. [Google Scholar] [CrossRef] [PubMed] - Van Wijk, B.C.; Stam, C.J.; Daffertshofer, A. Comparing brain networks of different size and connectivity density using graph theory. PLoS ONE
**2010**, 5, e13701. [Google Scholar] [CrossRef] [PubMed] - Kuyk, J.; Leijten, F.; Meinardi, H.; Spinhoven, P.; Van Dyck, R. The diagnosis of psychogenic non-epileptic seizures: A review. Seizure
**1997**, 6, 243–253. [Google Scholar] [CrossRef] [Green Version] - Lanius, U.F. Dissociation and endogenous opioids: A foundational role. In Neurobiology and Treatment of Traumatic Dissociation: Towards an Embodied Self; Springer: Berlin/Heidelberg, Germany, 2014. [Google Scholar]
- Szaflarski, J.P.; LaFrance, W.C., Jr. Psychogenic Nonepileptic Seizures (PNES) as a Network Disorder–Evidence from Neuroimaging of Functional (Psychogenic) Neurological Disorders. Epilepsy Curr.
**2018**, 18, 211–216. [Google Scholar] [CrossRef] [PubMed] - Kozlowska, K.; Chudleigh, C.; Cruz, C.; Lim, M.; McClure, G.; Savage, B.; Shah, U.; Cook, A.; Scher, S.; Carrive, P.; et al. Psychogenic non-epileptic seizures in children and adolescents: Part I–Diagnostic formulations. Clin. Child Psychol. Psychiatry
**2018**, 23, 140–159. [Google Scholar] [CrossRef] - Meppelink, A.M.; Pareés, I.; Beudel, M.; Little, S.; Yogarajah, M.; Sisodiya, S.; Edwards, M.J. Spectral power changes prior to psychogenic non-epileptic seizures: A pilot study. J. Neurol. Neurosurg. Psychiatry
**2017**, 88, 190–192. [Google Scholar] [CrossRef] - Arıkan, K.; Öksüz, Ö.; Metin, B.; Günver, G.; Laçin Çetin, H.; Esmeray, T.; Tarhan, N. Quantitative EEG findings in patients with psychogenic nonepileptic seizures. Clin. EEG Neurosci.
**2020**, 52, 175–180. [Google Scholar] [CrossRef] [PubMed] - Amiri, S.; Mirbagheri, M.M.; Asadi-Pooya, A.A.; Badragheh, F.; Zibadi, H.A.; Arbabi, M. Brain functional connectivity in individuals with psychogenic nonepileptic seizures (PNES): An application of graph theory. Epilepsy Behav.
**2021**, 114, 107565. [Google Scholar] [CrossRef] - Engel, J., Jr. A proposed diagnostic scheme for people with epileptic seizures and with epilepsy: Report of the ILAE Task Force on Classification and Terminology. Epilepsia
**2001**, 42, 796–803. [Google Scholar] [CrossRef] [PubMed] - Oldfield, R.C. The assessment and analysis of handedness: The Edinburgh inventory. Neuropsychologia
**1971**, 9, 97–113. [Google Scholar] [CrossRef] - Faul, F.; Erdfelder, E.; Buchner, A.; Lang, A.G. Statistical power analyses using G* Power 3.1: Tests for correlation and regression analyses. Behav. Res. Methods
**2009**, 41, 1149–1160. [Google Scholar] [CrossRef] [Green Version] - Boulila, W.; Ayadi, Z.; Farah, I.R. Sensitivity analysis approach to model epistemic and aleatory imperfection: Application to Land Cover Change prediction model. J. Comput. Sci.
**2017**, 23, 58–70. [Google Scholar] [CrossRef] - Delorme, A.; Makeig, S. EEGLAB: An open source toolbox for analysis of single-trial EEG dynamics including independent component analysis. J. Neurosci. Methods
**2004**, 134, 9–21. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Pion-Tonachini, L.; Kreutz-Delgado, K.; Makeig, S. The ICLabel dataset of electroencephalographic (EEG) independent component (IC) features. Data Brief
**2019**, 25, 104101. [Google Scholar] [CrossRef] [PubMed] - Welch, P. The use of fast Fourier transform for the estimation of power spectra: A method based on time averaging over short, modified periodograms. IEEE Trans. Audio Electroacoust.
**1967**, 15, 70–73. [Google Scholar] [CrossRef] [Green Version] - Klimesch, W. EEG alpha and theta oscillations reflect cognitive and memory performance: A review and analysis. Brain Res. Rev.
**1999**, 29, 169–195. [Google Scholar] [CrossRef] - Whitfield-Gabrieli, S.; Nieto-Castanon, A. Conn: A functional connectivity toolbox for correlated and anticorrelated brain networks. Brain Connect.
**2012**, 2, 125–141. [Google Scholar] [CrossRef] [Green Version] - Stam, C.J.; Nolte, G.; Daffertshofer, A. Phase lag index: Assessment of functional connectivity from multi channel EEG and MEG with diminished bias from common sources. Hum. Brain Mapp.
**2007**, 28, 1178–1193. [Google Scholar] [CrossRef] [PubMed] - Marzetti, L.; Basti, A.; Chella, F.; D’Andrea, A.; Syrjälä, J.; Pizzella, V. Brain functional connectivity through phase coupling of neuronal oscillations: A perspective from magnetoencephalography. Front. Neurosci.
**2019**, 13, 964. [Google Scholar] [CrossRef] [Green Version] - Sporns, O.; Honey, C.J. Small worlds inside big brains. Proc. Natl. Acad. Sci. USA
**2006**, 103, 19219–19220. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Latora, V.; Marchiori, M. Efficient behavior of small-world networks. Phys. Rev. Lett.
**2001**, 87, 198701. [Google Scholar] [CrossRef] [Green Version] - Watts, D.J.; Strogatz, S.H. Collective dynamics of ‘small-world’networks. Nature
**1998**, 393, 440. [Google Scholar] [CrossRef] - Fornito, A.; Zalesky, A.; Bullmore, E. Fundamentals of Brain Network Analysis; Academic Press: Cambridge, MA, USA, 2016. [Google Scholar]
- Tononi, G.; Sporns, O.; Edelman, G.M. A measure for brain complexity: Relating functional segregation and integration in the nervous system. Proc. Natl. Acad. Sci. USA
**1994**, 91, 5033–5037. [Google Scholar] [CrossRef] [Green Version] - Freeman, L.C. A set of measures of centrality based on betweenness. Sociometry
**1977**, 40, 35–41. [Google Scholar] [CrossRef] - Gogate, M.; Dashtipour, K.; Hussain, A. Visual Speech in Real Noisy Environments (VISION): A Novel Benchmark Dataset and Deep Learning-Based Baseline System. Interspeech
**2020**, 4521–4525. [Google Scholar] [CrossRef] - Gogate, M.; Dashtipour, K.; Adeel, A.; Hussain, A. CochleaNet: A robust language-independent audio-visual model for real-time speech enhancement. Inf. Fusion
**2020**, 63, 273–285. [Google Scholar] [CrossRef] - Rousseeuw, P.J.; Croux, C. Alternatives to the median absolute deviation. J. Am. Stat. Assoc.
**1993**, 88, 1273–1283. [Google Scholar] [CrossRef] - Vapnik, V.N. An overview of statistical learning theory. IEEE Trans. Neural Netw.
**1999**, 10, 988–999. [Google Scholar] [CrossRef] [Green Version] - Safaei, M.; Ismail, A.S.; Chizari, H.; Driss, M.; Boulila, W.; Asadi, S.; Safaei, M. Standalone noise and anomaly detection in wireless sensor networks: A novel time-series and adaptive Bayesian-network-based approach. Softw. Pract. Exp.
**2020**, 50, 428–446. [Google Scholar] [CrossRef] - Steinwart, I.; Christmann, A. Support Vector Machines; Springer Science & Business Media: Berlin/Heidelberg, Germany, 2008. [Google Scholar]
- Izenman, A.J. Linear discriminant analysis. In Modern Multivariate Statistical Techniques; Springer: Berlin/Heidelberg, Germany, 2013; pp. 237–280. [Google Scholar]
- Yegnanarayana, B. Artificial Neural Networks; PHI Learning Pvt. Ltd.: Delhi, India, 2009. [Google Scholar]
- Boulila, W. A top-down approach for semantic segmentation of big remote sensing images. Earth Sci. Inform.
**2019**, 12, 295–306. [Google Scholar] [CrossRef] - Achard, S.; Bullmore, E. Efficiency and cost of economical brain functional networks. PLoS Comput. Biol.
**2007**, 3, e17. [Google Scholar] [CrossRef] [PubMed] - Fries, P. A mechanism for cognitive dynamics: Neuronal communication through neuronal coherence. Trends Cogn. Sci.
**2005**, 9, 474–480. [Google Scholar] [CrossRef] - Thut, G.; Miniussi, C.; Gross, J. The functional importance of rhythmic activity in the brain. Curr. Biol.
**2012**, 22, R658–R663. [Google Scholar] [CrossRef] [Green Version] - Jeong, J. EEG dynamics in patients with Alzheimer’s disease. Clin. Neurophysiol.
**2004**, 115, 1490–1505. [Google Scholar] [CrossRef] [PubMed] - Richardson, M.P. Large scale brain models of epilepsy: Dynamics meets connectomics. J. Neurol. Neurosurg. Psychiatry
**2012**, 83, 1238–1248. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Halgren, M.; Ulbert, I.; Bastuji, H.; Fabó, D.; Eross, L.; Rey, M.; Devinsky, O.; Doyle, W.K.; Mak-McCully, R.; Halgren, E.; et al. The generation and propagation of the human alpha rhythm. bioRxiv
**2018**, 202564. [Google Scholar] [CrossRef] [PubMed] - Devinsky, O.; Gazzola, D.; LaFrance, W.C., Jr. Differentiating between nonepileptic and epileptic seizures. Nat. Rev. Neurol.
**2011**, 7, 210. [Google Scholar] [CrossRef] [PubMed] - Reuber, M.; Fernandez, G.; Helmstaedter, C.; Qurishi, A.; Elger, C. Evidence of brain abnormality in patients with psychogenic nonepileptic seizures. Epilepsy Behav.
**2002**, 3, 249–254. [Google Scholar] [CrossRef] - Fox, M.D.; Raichle, M.E. Spontaneous fluctuations in brain activity observed with functional magnetic resonance imaging. Nat. Rev. Neurosci.
**2007**, 8, 700–711. [Google Scholar] [CrossRef] - Hernando, K.A.; Szaflarski, J.P.; Ver Hoef, L.W.; Lee, S.; Allendorfer, J.B. Uncinate fasciculus connectivity in patients with psychogenic nonepileptic seizures: A preliminary diffusion tensor tractography study. Epilepsy Behav.
**2015**, 45, 68–73. [Google Scholar] [CrossRef] - Devinsky, O.; Mesad, S.; Alper, K. Nondominant hemisphere lesions and conversion nonepileptic seizures. J. Neuropsychiatry Clin. Neurosci.
**2001**, 13, 367–373. [Google Scholar] [CrossRef] [PubMed] - Soriano, M.C.; Niso, G.; Clements, J.; Ortín, S.; Carrasco, S.; Gudín, M.; Mirasso, C.R.; Pereda, E. Automated detection of epileptic biomarkers in resting-state interictal MEG data. Front. Neuroinform.
**2017**, 11, 43. [Google Scholar] [CrossRef] [PubMed] - Varone, G.; Hussain, Z.; Sheikh, Z.; Howard, A.; Boulila, W.; Mahmud, M.; Howard, N.; Morabito, F.C.; Hussain, A. Real-Time Artifacts Reduction during TMS-EEG Co-Registration: A Comprehensive Review on Technologies and Procedures. Sensors
**2021**, 21, 637. [Google Scholar] [CrossRef] [PubMed]

**Figure 1.**Flowchart of the proposed methodology. Panel (

**A**) shows the international standardized 19 scalp EEG channels location. Panel (

**B**) depicts the signal preprocessing pipeline used to clean the collected EEG time series. Panel (

**C**) highlights the power spectral density processing, where each single EEG channel was fed in the pwelch function. Panel (

**D**) depicts the EEG-based PLI network analysis. At this stage, each PLI matrix was thresholded and the adjacency matrix was computed. Panel (

**E**) lists the indices used to test the performance of the network. In detail, we used global efficiency, node betweenness, cluster coefficient, small-worldness, and shortest path length, such as network indices. Panel (

**F**) highlights a schematic representation of three ML classifiers, respectively, support vector machine (SVM), linear discriminant analysis (LDA), and Multilayer perceptron (MLP).

**Figure 2.**MLP architecture. Our MLP architecture comprises one hidden layer with 18 units, followed by a softmax layer, for binary classification.

**Figure 3.**Box plots of normalized PSD values between pairwise of similar sensors in PNES versus HC. Panel (

**A**) depict the PSD results of pairwise comparison in delta band, panel (

**B**) in theta, (

**C**) in alpha, and (

**D**) in beta band.

**Figure 4.**Relative normalized PSD, averaged over the 19 electrodes, in four frequency bands for frontal (see panel (

**A**)), central (see panel (

**B**)), parieto-occipital area (see panel (

**C**)) between PNES and the control group. The horizontal mark within each box represents the median, the edges of the box represent the first and third quartile. The horizontal line denotes significant difference between groups, where ** stand for p < 0.05, and *** for p < 0.001. The PSD measures was computed in PNES and control in the delta, theta, alpha, and beta frequency bands.

**Figure 5.**Distribution of PLI for th = 0.05 level of the connection strength, in healthy subjects and PNES in delta, theta, alpha, and beta frequency bands. The line thickness highlights path with high value of PLI.

**Figure 6.**Distribution of PLI for th = 0.15 level of the connection strength, in healthy subjects and PNES in delta, theta, alpha, and beta frequency bands.

**Figure 7.**Distribution of PLI for th = 0.25 level of the connection strength, in healthy subjects and PNES in delta, theta, alpha, and beta frequency bands.

**Figure 8.**Box plots of network coefficients measured in PNES and HC. Here, we report normalized network coefficients values of: CC, Ge, NB, SPL, and SW in delta, theta, alpha, and beta bands. The horizontal mark within each box represents the median, the edges of the box represent the first and third quartile, and the whiskers extend to the most extreme data points that are not considered outliers, where the symbols ** stand for p < 0.05, and *** for p < 0.001. The PSD measures was computed in PNES and control in the delta, theta, alpha, and beta frequency bands.

**Figure 9.**ROC curves of the proposed SVM, LDA, and MLP classifiers for binary classification task (PNES versus HC).

**Table 1.**The table report results from a statistical analysis across cluster of EEG sensors (see EEG layout parcelization in Section 5.1) in all of the frequencies band under analysis. We evaluated Global efficiency, Node betweenness, Cluster coefficient, Small world, and Shortest path characteristic. We performed a post-hoc analysis for multi-comparison. The p-value (p < 0.05) corresponds to significant difference in network index values for the testing conditions. See Section 3.7 for further information.

Frontal Area | Central Area | Parieto-Occipital Area | ||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|

Graph Index | δ | θ | α | β | δ | θ | α | β | δ | θ | α | β |

Global efficiency | 0.0532 | 0.132 | 0.102 | 0.251 | 0.04 | 0.06 | 0.0128 | 0.0013 | 0.073 | 0.122 | 0.016 | 0.00123 |

Node betweenness | 0.273 | 0.322 | 0.421 | 0.151 | 0.073 | 0.122 | 0.033 | 0.023 | 0.013 | 0.018 | 0.263 | 0.032 |

Cluster coefficient | 0.074 | 0.126 | 0.452 | 0.321 | 0.142 | 0.785 | 0.0412 | 0.022 | 0.561 | 0.174 | 0.0430 | 0.0012 |

Small world | 0.752 | 0.134 | 0.144 | 0.434 | 0.33 | 0.431 | 0.04 | 0.014 | 0.335 | 0.453 | 0.034 | 0.174 |

Shortest path | 0.331 | 0.123 | 0.424 | 0.041 | 0.021 | 0.014 | 0.143 | 0.041 | 0.012 | 0.033 | 0.012 | 0.044 |

**Table 2.**Binary classification performance (Precision, Recall, F1 score, and Accuracy) of SVM, MLP, and LDA.

Precision | SVM | MLP | LDA |
---|---|---|---|

PNES | 77.74% | 78.23% | 75.42% |

HC | 89.22% | 91.23% | 93.42% |

AVG | 83.48% | 85.73% | 84.42% |

Recall | |||

PNES | 95.24% | 96.73% | 96.42% |

HC | 54.22% | 76.42% | 63.12% |

AVG | 74.73% | 86.57% | 79.77 |

F1 score | |||

PNES | 87.44% | 86.75% | 83.26% |

HC | 65.61% | 71.22% | 65.07% |

AVG | 76.52% | 78.98% | 74.16% |

Accuracy | |||

PNES | 89.21% | 96.82% | 82.16% |

HC | 69.61% | 85.22% | 71.27% |

AVG | 79.41% | 91.02% | 76.72% |

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |

© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Varone, G.; Boulila, W.; Lo Giudice, M.; Benjdira, B.; Mammone, N.; Ieracitano, C.; Dashtipour, K.; Neri, S.; Gasparini, S.; Morabito, F.C.;
et al. A Machine Learning Approach Involving Functional Connectivity Features to Classify Rest-EEG Psychogenic Non-Epileptic Seizures from Healthy Controls. *Sensors* **2022**, *22*, 129.
https://doi.org/10.3390/s22010129

**AMA Style**

Varone G, Boulila W, Lo Giudice M, Benjdira B, Mammone N, Ieracitano C, Dashtipour K, Neri S, Gasparini S, Morabito FC,
et al. A Machine Learning Approach Involving Functional Connectivity Features to Classify Rest-EEG Psychogenic Non-Epileptic Seizures from Healthy Controls. *Sensors*. 2022; 22(1):129.
https://doi.org/10.3390/s22010129

**Chicago/Turabian Style**

Varone, Giuseppe, Wadii Boulila, Michele Lo Giudice, Bilel Benjdira, Nadia Mammone, Cosimo Ieracitano, Kia Dashtipour, Sabrina Neri, Sara Gasparini, Francesco Carlo Morabito,
and et al. 2022. "A Machine Learning Approach Involving Functional Connectivity Features to Classify Rest-EEG Psychogenic Non-Epileptic Seizures from Healthy Controls" *Sensors* 22, no. 1: 129.
https://doi.org/10.3390/s22010129