Torse Dattaprasad, Desai Veena, Khanai Rajashri. An optimized design of seizure detection system using joint feature extraction of multichannel EEG signals[J]. The Journal of Biomedical Research, 2020, 34(3): 191-204. DOI: 10.7555/JBR.33.20190019
Citation:
Torse Dattaprasad, Desai Veena, Khanai Rajashri. An optimized design of seizure detection system using joint feature extraction of multichannel EEG signals[J]. The Journal of Biomedical Research, 2020, 34(3): 191-204. DOI: 10.7555/JBR.33.20190019
Torse Dattaprasad, Desai Veena, Khanai Rajashri. An optimized design of seizure detection system using joint feature extraction of multichannel EEG signals[J]. The Journal of Biomedical Research, 2020, 34(3): 191-204. DOI: 10.7555/JBR.33.20190019
Citation:
Torse Dattaprasad, Desai Veena, Khanai Rajashri. An optimized design of seizure detection system using joint feature extraction of multichannel EEG signals[J]. The Journal of Biomedical Research, 2020, 34(3): 191-204. DOI: 10.7555/JBR.33.20190019
Department of Electronics and Communication Engineering, KLS Gogte Institute of Technology, Belagavi, Karnataka 590008, India
2.
Department of Electronics and Communication Engineering, KLE Dr. M. S. Sheshgiri College of Engineering and Technology, Belagavi, Karnataka 590008, India
Dattaprasad Torse, Department of Electronics and Communication Engineering, KLS Gogte Institute of Technology, Udyambag, Khanapur road, Belagavi, Karnataka590008, India. Tel/Fax: +91-9480113464/+91-831-2441909, E-mail: datorse@git.edu
The detection of seizure onset and events using electroencephalogram (EEG) signals are important tasks in epilepsy research. The literature available on seizure detection has discussed the implementation of advanced signal processing algorithms using tools accessed over the cloud. However, seizure monitoring application needs near sensor processing due to privacy and latency issues. In this paper, a real time seizure detection system has been implemented using an embedded system. The proposed system is based on ensemble empirical mode decomposition (EEMD) and tunable-Q wavelet transform (TQWT) algorithms. The analysis and classification of non-stationary EEG signals require the wavelet transform with high Q-factor. However, direct use of TQWT increases the computational complexity of feature extraction from multivariate EEG signals. In this paper, the first step is to process the signal by using EEMD to obtain 8 intrinsic mode functions (IMFs). The Kraskov (KraEn), sample (SampEn), and permutation (PermEn) entropy features of IMFs are extracted and based on optimum values, and 4 IMFs are decomposed using TQWT. Secondly, centered correntropy (CenCorrEn) features of the 1st and 16th sub-band of TQWT have been used as classifier inputs. The performance of multilayer perceptron neural networks (MLPNN), least squares support vector machine (LSSVM), and random forest (RF) classifiers has been tested on the multichannel EEG data recorded from a local hospital. The RF classifier has produced the highest accuracy of 96.2% in classifying the signals. The proposed scheme has been employed in developing an embedded seizure detection system to assist neurologists in making seizure diagnostic decisions.
Observational studies in epidemiology have identified a correlation between hypothyroidism and cholelithiasis[1-2]. However, the causal relationship between the two diseases remains unclear. To investigate the potential causal relationship, we employed a two-sample bidirectional Mendelian randomization (MR) analysis.
The data for hypothyroidism was obtained from the IEU Open GWAS Project (https://gwas.mrcieu.ac.uk, ukb-a-77, N = 337 159), and the data for cholelithiasis were sourced from the FinnGen biobank (https://r8.finngen.fi/, finngen_R8_K11_CHOLELITH, N = 334 277).
The detailed process of this study is shown in Supplementary Fig. 1 (available online). We used multiple analytical methods, including inverse variance weighted (IVW), MR Egger, weighted median, simple mode, and weighted mode, alonside various sensitivity analyses. To reinforce our findings, we exchanged the exposure and outcome data and subsequently determined the causal effect of cholelithiasis on hypothyroidism following the same procedure.
Initially, a total of 83 single nucleotide polymorphisms (SNPs) were screened from the exposure-genome-wide association study (GWAS) dataset (P < 5e-8, r2 > 0.001, 10 000 kb). To avoid the weak tool bias, only 69 SNPs (F-statistic > 10) were included in the analysis. We also extracted relevant information for these 69 SNPs from the outcome GWAS dataset, and 66 SNPs were identified. For SNPs missing in the outcome dataset, we used the online tool LDLink (https://ldlink.nih.gov/) to find proxy SNPs (LD r2 > 0.8, present in the outcome dataset, and with consistent alleles). SNPs without suitable proxies were excluded[3]. A secondary selection of SNPs was performed after data harmonization. We removed the SNPs that had palindromic structures and those marked as "MR-Keep = False" (rs2823272, rs2921053, rs66749983, rs7582694, rs761357, and rs7768019). Through PhenoScanner V2 screening (http://www.phenoscanner.medschl.cam.ac.uk), we excluded SNPs associated with cholelithiasis or known risk factors, including primary sclerosing cholangitis (rs4276275 and rs7090530), body mass index (rs705702), diabetes-related traits (rs12980063, rs229540, rs3184504, and rs3850765), chronic liver disease and rheumatoid arthritis (rs11052877). We found that all SNPs primarily affected the exposure rather than the outcome by the Steiger directionality test. Finally, 52 independent and effective SNPs were selected for analysis (Supplementary Data 1, available online).
After rigorous MR analyses, we found that hypothyroidism has a strong causal effect on cholelithiasis: MR-Egger (OR 4.067, 95% CI 1.175–14.082, P = 0.031), weighted median (OR 3.841, 95% CI 1.667–8.852, P = 0.002), IVW (OR 2.950, 95% CI 1.663–5.233, P < 0.001), and weighted mode (OR 2.839, 95% CI 1.136–7.096, P = 0.030). However, no significant result was found using the simple mode (P = 0.342). IVW is the most important analysis method in MR studies, because it provides an unbiased causality estimation in an ideal state[4]. Although MR-Egger may infer the corrected causality, its statistical efficiency is lower than that of IVW[5]. When more than 50% of SNPs are effective, the weighted median results are accurate[6]. Although the simple mode provides robustness for pleiotropy, it is not as powerful as IVW[7]. The weighted mode is sensitive to the difficult bandwidth selection for mode estimation. When the maximal subset of tools with similar causal effects is valid, the weighted mode results are reliable[8]. According to the IVW result, individuals with hypothyroidism-related traits had a 2.95-fold increased risk of developing cholelithiasis compared with those without hypothyroidism.
We then performed multiple sensitivity analyses to verify the robustness of our results. The result of heterogeneity test was not statistically significant (P > 0.05). The multiplicative random effects mode showed a significant association between hypothyroidism and cholelithiasis (P < 0.05, b > 0). The MR-Egger intercept test (P = 0.57) indicated no significant pleiotropic effect of the SNPs. Additionally, we performed the MR-PRESSO test and found no significant outliers.
The scatter plot showed that an increased risk of cholelithiasis was associated with hypothyroidism (Fig. 1A). The red line at the bottom of the forest plot indicated that hypothyroidism increased the risk of cholelithiasis in the IVW (Fig. 1C). The leave-one-out test showed that all the lines were on the right side of 0, demonstrating the robustness of our results (Fig. 1D). The funnel plot was symmetrical, consistent with the results of the heterogeneity test (Fig. 1E).
A: Scatter plot illustrating the distribution of individual ratio estimates of hypothyroidism with cholelithiasis as the outcome. B: Scatter plot illustrating the distribution of individual ratio estimates of cholelithiasis with hypothyroidism as the outcome. C: Forest plot. Each horizontal solid line reflects the result estimated by a single SNP in the Wald ratio method. The lowest red line indicates that hypothyroidism increases the risk of cholelithiasis under IVW. D: Leave-one-out sensitivity analysis of the causal effect of hypothyroidism on cholelithiasis. The red line indicates reliable estimations from the IVW and MR-Egger methods. E: Funnel plot. The points on both sides are roughly symmetrically distributed. F: Four mechanisms for hypothyroidism increasing the risk of cholelithiasis.
We exchanged the exposure and outcome data to determine the causal effect of cholelithiasis on risk of hypothyroidism. Following the same steps, a total of 56 SNPs were screened, and only 47 SNPs (F-statistic > 10) were included in the current study. After data harmonization and removal of nine SNPs marked as "MR-Keep = False", 38 SNPs were selected. Through PhenoScanner V2 screening, rs174592 (hypothyroidism-related) was removed. Finally, 37 SNPs were selected (Supplementary Data 2, available online). The results of MR-Egger (OR 1.000, 95% CI 0.996–1.005, P = 0.948), weighted median (OR 1.001, 95% CI 0.998–1.004, P = 0.623), IVW (OR 1.002, 95% CI 0.999–1.005, P = 0.054), simple mode (OR 1.000, 95% CI 0.994–1.007, P = 0.978), and weighted mode (OR 0.999, 95% CI 0.996–1.003, P = 0.767) were analyzed. Because the IVW result was not significant, and the directions of the other methods were inconsistent, the MR analyses of cholelithiasis affecting hypothyroidism were not significant (Fig. 1B).
The association between hypothyroidism and cholelithiasis has been controversial for decades. Although numerous observational studies have established a association between the two diseases, which is consistent with our findings[9–10], hypothyroidism increasing the risk of cholelithiasis may be related to several mechanisms, including disorders of lipid metabolism, decreased gallbladder motor function, impaired relaxation of the human sphincter of Oddi, and reduced bile secretion (Fig. 1F).
First, increasing evidence has demonstrated a strong connection between hypothyroidism and lipid metabolic disorders. Hypothyroidism significantly increased the expression of the ABCG8 gene in the liver[10], promoting the entry of more cholesterol into bile and directly contributing to cholesterol gallstone formation[11]. Moreover, low levels of thyroid hormones decreased the expression levels of hepatic low-density lipoprotein receptors, limiting cholesterol uptake by hepatocytes and resulting in hypercholesterolemia[12]. Cholesterol 7α-hydroxylase (CYP7A1) is the rate-limiting enzyme in the liver, which converts cholesterol into primary bile acids. Studies have shown that thyroid hormones may increase CYP7A1 mRNA expression levels, which is crucial because bile acid synthesis is an important mechanism for the liver to regulate excess cholesterol[13]. Interestingly, hypothyroid patients often exhibit elevated thyroid-stimulating hormone (TSH) levels, which have dual effects on cholesterol metabolism. Elevated TSH levels not only increase serum cholesterol by upregulating HMG-CoA reductase (the rate-limiting enzyme in cholesterol synthesis), but also inhibit the conversion of cholesterol to bile acids through the SREBP-2/HNF-4a/CYP7A1 pathway[14]. In addition, the effect of hypothyroidism on renal circulation is noteworthy, as it reduces renal plasma flow and lowers the glomerular filtration rate, which may indirectly influence lipid metabolism[12].
Second, hypothyroidism affects gallbladder motility, although the specific molecular mechanisms remain unclear. The reduced metabolic capacity in hypothyroid patients may impair gallbladder contraction. Additionally, hypercholesterolemia caused by hypothyroidism increases cholesterol levels within gallbladder smooth muscle cells, thereby affecting smooth muscle contraction[15]. These factors together compromise efficient gallbladder emptying, potentially exacerbating biliary complications.
Third, hypothyroid patients exhibit a decreased ability of the Oddi sphincter to regulate bile excretion, increasing the risk of common bile duct stones. This occurs because the relaxation of the Oddi sphincter is partially dependent on thyroid hormone action. In Oddi sphincter cells, thyroid hormones cause cell membrane hyperpolarization by opening adenosine triphosphate (ATP)-sensitive K+ channels, reducing calcium ion influx and limiting smooth muscle contraction[16]. Therefore, impaired relaxation of the Oddi sphincter further disrupts bile flow.
Lastly, hypothyroidism reduces bile secretion by hepatocytes, diminishing the ability to clear biliary sediment[2]. Since bile acids are a critical component of bile, we hypothesize that disturbances in bile acid metabolism play a significant role in the reduced bile secretion observed in hypothyroidism. Beyond the previously noted decrease in bile acid synthesis, the impairment in bile acid reutilization also warrants attention. Notably, bile acid transporter expression is significantly reduced in the livers of animals with hypothyroidism, disrupting the enterohepatic circulation of bile acids[12]. Furthermore, a recent study has demonstrated that hypothyroidism increases the incidence of cholesterol stones in mice by enhancing the hydrophobicity of primary bile acids[17]. Despite these findings, the underlying biological mechanisms require to be fully elucidated.
The main advantage of the current study is that we used two-sample bidirectional MR analyses. Because SNPs are theoretically randomly distributed, this significantly reduces the effects of confounding factors and reverse causality[18]. In addition, the data of the current study came from a collection of multiple studies, and the sample size was significantly larger than in any previous observational study, which should increase the accuracy of the results.
However, the current study also has some limitations. First, the individuals we studied were exclusively Europeans, and the findings may not be applicable to the general populations. Second, because we employed several methodologies, it is likely to have some potential multiple-test effects. Nonetheless, the results of sensitivity analyses demonstrated the reliability of our findings. Third, the current study may only provide a preliminary assessment of the causal relationship between hypothyroidism and cholelithiasis, and further prospective studies are necessary to validate our findings.
The current study found a strong causal effect of hypothyroidism on cholelithiasis but did not find a similar effect in the opposite direction. The study helps remind clinicians to pay more attention to the link between the two diseases.
The study was sponsored by grants from the Jiangsu Province 333 High-level Talent Training Project (Grant No. LGY2016010), the Nanjing Science and Technology Development Plan (Grant No. 201715003), and the Jiangsu Province Six Talent Peaks (Grant No. WSN-030).
We thank the FinnGen biobank and the UK Biobank for providing GWAS data.
Yours Sincerely,Xu Han, Hong Zhu✉ Department of Gastroenterology, the First Affiliated Hospital of Nanjing Medical University,Nanjing, Jiangsu 210029, China.✉Corresponding author: Hong Zhu. E-mail: zhuhong1059@126.com.
Devinsky, Orrin, and J. Kiffin Penry. Quality of life in epilepsy: the clinician's view[J]. Epilepsia, 1993, 34: S4–S7.
[2]
Panayiotopoulos CP. A clinical guide to epileptic syndromes and their treatment[M]. 2nd ed. London: Springer, 2007: 185–206.
[3]
WHO. Mental health: new understanding, new hope[R]. Geneva, Switzerland: WHO, 2001.
[4]
Kumar S, Madaan R, Bansal G, et al. Plants and plant products with potential anticonvulsant activity—a review[J]. Pharmacog Commun, 2012, 2(1): 3–99.
[5]
Gargiulo G, Calvo RA, Bifulco P, et al. A new EEG recording system for passive dry electrodes[J]. Clin Neurophysiol, 2010, 121(5): 686–693. doi: 10.1016/j.clinph.2009.12.025
[6]
Siegel AM. Presurgical evaluation and surgical treatment of medically refractory epilepsy[J]. Neurosurg Rev, 2004, 27(1): 1–18. doi: 10.1007/s10143-003-0305-6
[7]
Giannakakis G, Sakkalis V, Pediaditis M, et al. Methods for seizure detection and prediction: an overview[M]//Sakkalis V. Modern Electroencephalographic Assessment Techniques: Theory and Applications. New York, NY: Humana Press, 2014: 131–157.
[8]
Verma N, Shoeb A, Bohorquez J, et al. A micro-power EEG acquisition SoC with integrated feature extraction processor for a chronic seizure detection system[J]. IEEE J Solid-State Circuits, 2010, 45(4): 804–816. doi: 10.1109/JSSC.2010.2042245
[9]
Hassan AR, Siuly S, Zhang YC. Epileptic seizure detection in EEG signals using tunable-Q factor wavelet transform and bootstrap aggregating[J]. Comput Methods Programs Biomed, 2016, 137: 247–259.
[10]
Hassan AR, Bhuiyan MIH. Automatic sleep scoring using statistical features in the EMD domain and ensemble methods[J]. Biocybern Biomed Eng, 2016, 36(1): 248–255.
[11]
Hassan AR, Bhuiyan MIH. An automated method for sleep staging from EEG signals using normal inverse Gaussian parameters and adaptive boosting[J]. Neurocomputing, 2017, 219: 76–87.
[12]
Hassan AR, Bhuiyan MIH. A decision support system for automatic sleep staging from EEG signals using tunable Q-factor wavelet transform and spectral features[J]. J Neurosci Methods, 2016, 271: 107–118.
[13]
Hassan AR, Bashar SK, Bhuiyan MIH. Automatic classification of sleep stages from single-channel electroencephalogram[C]//Proceedings of 2015 Annual IEEE India Conference. New Delhi, India: IEEE, 2015.
[14]
Hassan AR, Huda MN, Sarker F, et al. An overview of brain machine interface research in developing countries: opportunities and challenges[C]//Proceedings of the 2016 5th International Conference on Informatics, Electronics and Vision. Dhaka, Bangladesh: IEEE, 2016.
[15]
Rahman M, Bhuiyan MIH, Hassan AR. Sleep stage classification using single-channel EOG[J]. Comput Biol Med, 2018, 102: 211–220. doi: 10.1016/j.compbiomed.2018.08.022
[16]
Hassan AR, Subasi A. A decision support system for automated identification of sleep stages from single-channel EEG signals[J]. Knowl-Based Syst, 2017, 128: 115–124.
[17]
Kohavi R, John GH. Wrappers for feature subset selection[J]. Artif Intell, 1997, 97(1-2): 273–324. doi: 10.1016/S0004-3702(97)00043-X
[18]
Huang NE, Shen Z, Long SR, et al. The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis[J]. Proc Roy Soc A Math Phys Eng Sci, 1998, 454(1971): 903–995. doi: 10.1098/rspa.1998.0193
[19]
Rato RT, Ortigueira MD, Batista AG. On the HHT, its problems, and some solutions[J]. Mech Syst Signal Process, 2008, 22(6): 1374–1394. doi: 10.1016/j.ymssp.2007.11.028
[20]
Wu ZH, Huang NE. Ensemble empirical mode decomposition: a noise-assisted data analysis method[J]. Adv Adapt Data Anal, 2009, 1(1): 1–41.
[21]
Acharya UR, Molinari F, Sree SV, et al. Automated diagnosis of epileptic EEG using entropies[J]. Biomed Signal Process Control, 2012, 7(4): 401–408. doi: 10.1016/j.bspc.2011.07.007
[22]
Kraskov A, Stögbauer H, Grassberger P. Estimating mutual information[J]. Phys Rev E Stat Nonlin Soft Matter Phys, 2004, 69(6): 066138. doi: 10.1103/PhysRevE.69.066138
[23]
Riedl M, Müller A, Wessel N. Practical considerations of permutation entropy: a tutorial review[J]. Eur Phys J Spec Top, 2013, 222(2): 249–262.
[24]
Hu M, Liang HL. Intrinsic mode entropy based on multivariate empirical mode decomposition and its application to neural data analysis[J]. Cogn Neurodyn, 2011, 5(3): 277–284. doi: 10.1007/s11571-011-9159-8
[25]
Liu WF, Pokharel P, Xu JW, et al. Correntropy for random variables: properties and applications in statistical inference[M]//Principe JC. Information Theoretic Learning: Renyi's Entropy and Kernel Perspectives. New York: Springer, 2010: 385–413.
[26]
Haykin S, Network N. A comprehensive foundation[J]. Neural Netw, 2004, 2: 41.
[27]
Sharma M, Pachori RB, Acharya UR. A new approach to characterize epileptic seizures using analytic time-frequency flexible wavelet transform and fractal dimension[J]. Pattern Recognit Lett, 2017, 94: 172–179.
[28]
Suykens JAK, Vandewalle J. Least squares support vector machine classifiers[J]. Neural Process Lett, 1999, 9(3): 293–300. doi: 10.1023/A:1018628609742
[29]
Breiman L. Random forests[J]. Mach Learn, 2001, 45(1): 5–32.
[30]
Sharma R, Pachori RB, Acharya UR. Application of entropy measures on intrinsic mode functions for the automated identification of focal electroencephalogram signals[J]. Entropy, 2015, 17(2): 669–691. doi: 10.3390/e17020669
[31]
Haynes W. Student's t-test[M]//Dubitzky W, Wolkenhauer O, Cho KH, et al. Encyclopedia of Systems Biology. New York: Springer, 2013: 2023–2025.
[32]
Gupta V, Priya T, Yadav AK, et al. Automated detection of focal EEG signals using features extracted from flexible analytic wavelet transform[J]. Pattern Recognit Lett, 2017, 94: 180–188. doi: 10.1016/j.patrec.2017.03.017
Authors and Reviewers
DownLoad:
