|Year : 2018 | Volume
| Issue : 1 | Page : 61-65
Time-domain heart rate variability-based computer-aided prognosis of lung cancer
Reema Shyamsunder Shukla, Yogender Aggarwal
Department of Bio-Engineering, Birla Institute of Technology, Mesra, Ranchi, Jharkhand, India
|Date of Web Publication||23-Aug-2018|
Ms. Reema Shyamsunder Shukla
Department of Bio-Engineering, Birla Institute of Technology, Mesra, Ranchi, Jharkhand
Source of Support: None, Conflict of Interest: None
Objective: Incidence of lung cancer (LC) is increasing day by day with exposure to smoke, radiation, and chemicals; LC is one of the leading causes of death. The major difficulty in treatment was delayed diagnosis. This study aims to propose a time-domain heart rate variability (HRV) feature-based automated system in LC prediction and its staging. Materials and Methods: HRV analysis was done using recorded electrocardiographic signal from 104 LC participants and 30 control volunteers. Artificial neural network (ANN) and support vector machine (SVM) were implemented on HRV time-domain features for early prognosis of the disorder. Statistical significance of HRV parameters was tested, and graphical user interface (GUI) was also implemented. Results: It was revealed that progression of cancer causes low HRV. An accuracy of 89.64% and 100% was obtained with ANN and SVM, respectively, in automated cancer prediction. Statistical analysis suggested the significance of data at P < 0.05 between different performance statuses among patients. Conclusion: The severity of LC alters the sympathovagal balance through autonomic dysfunction. HRV analysis with an expert system was found useful for the early diagnosis of the disease, and thus, a noninvasive technique is of prognostic importance in classifying LC stages. The GUI designed for clinicians can help them to diagnose the Eastern Cooperative Oncology Group performance status scale of future patients.
Keywords: Artificial neural network, autonomic dysfunction, heart rate variability, lung cancer, support vector machine
|How to cite this article:|
Shukla RS, Aggarwal Y. Time-domain heart rate variability-based computer-aided prognosis of lung cancer. Indian J Cancer 2018;55:61-5
| » Introduction|| |
Incidence of cancer is increasing day by day despite development in medical science; cancer has been diagnosed to be the leading cause of death. Lung cancer (LC) is one of the deadly diseases across the globe. It can be categorized into benign, potentially malignant (precancer), or malignant (cancer). LC mainly arises in the epithelial lining of the bronchial tree and originates in the main and lobar bronchi. Squamous cell carcinoma, adenocarcinoma, and small cell carcinoma are different histological types of LC. The deposition of inhaled particles mainly due to smoking and tobacco consumption in the upper bronchial tree may result into an outgrowth of LC and are major causes of mortality among patients, generally in men.,,, However, the major hurdle in treatment is delayed diagnosis.
A review of literature revealed an association of cancer with autonomic nervous system (ANS) dysfunction with reduced heart rate variability (HRV) probably due to chronic stress.,,,, HRV can be defined as the variation in heart rate evaluated as beat-to-beat intervals in time domain. HRV of recorded electrocardiogram (ECG) describes the activity of both sympathetic (SNS) and parasympathetic (PNS) part of the ANS., The alterations in ANS tone regulates the cardiovascular activity. Few literature also observed the prognostic role of few time-domain parameters in the identification of cancer patients and their survival.,, Further, literature also suggested the application of HRV spectral analysis in cancer prediction., Apart from cancer, HRV analysis was successfully used to evaluate number of diseases of cardiovascular origin and others.,,
In the past years, much focus has been given to automated identification of cancer. The computer-aided early prognosis of disease may improve the survival rate to 50% with an additional aid to clinicians. However, most studies incorporated medical images and clinicopathological parameters to classify and predict the disease with an accuracy of 97% with support vector machine (SVM) and 96.04% with artificial neural network (ANN). However, no work was reported to the best of authors' knowledge in HRV-based identification and cancer staging despite correlation between cancer severity with ANS. Thus, this study aims to identify the prognostic value of time-domain HRV parameters in LC prediction with the development of automated system.
| » Materials and Methods|| |
Subjects and biosignal acquisition
Lead II ECG filtered at 0.5–35 Hz was recorded (using Ag/AgCl disposable electrode and MP45 bioamplifier [Biopac Systems Inc., USA]) and sampled at 200 samples/second for 5 min in supine position at room temperature of 25°C from 104 (6 were the Eastern Cooperative Oncology Group [ECOG1], 8 ECOG2, 30 ECOG3, and 60 ECOG4) LC patients along with 30 controls. Initially, 110 LC participants were considered in which 11 participants were excluded from the study, 5 had diabetes, 2 hypertension, 2 cardiac history, and 2 atrial fibrillation. The age group in LC was ECOG1 55–80 years (median age 68 years), ECOG2 55–70 years (median age 63 years), ECOG3 35–80 years (median age 58 years), and ECOG 4 35–75 years (median age 55 years). Controls were in the age group of 25–70 years (median age 48 years). The patients and controls were made to rest 30 min before the recording took place. The recording was performed in accordance with the ethical standards of the Declaration of Helsinki, and a written consent has been obtained from the patients and volunteers. The tachogram was obtained from recorded signal with R-wave threshold level set to 0.5, and interpolation was performed at cubic-spline frequency of 8 Hz (Acknowledge 4.0 (Biopac Systems Inc., USA)). HRV time-domain features were extracted from obtained tachogram using HRV analysis tool (Kubios HRV 2.0, Finland). The features include mean RR interval (mRR), mean heart rate (mHR), standard deviation of normal-to-normal (NN) interval (SDNN), square root of the mean squared differences of successive NN interval (RMSSD), HRV triangular index (TI), triangular interpolation of NN intervals (TiNN), standard deviation of heart rate (SDHR), number of successive NN interval >50 ms (NN50), and proportion derived by dividing NN50 by the total number of NN intervals (pNN50). The details of HRV analysis have been discussed earlier, which was also used for this study. The ECOG performance status (PS) scale is explained in detail in our previous study.
The soft computing techniques, ANN and SVM, were employed for classification. The architecture of 9:100:5 and 6:100:5 was trained and tested with HRV features. ANN using feedforward network and SVM using radial basis function kernel was used as classifiers. ANN was implemented using Levenberg–Marquardt algorithm in MATLAB 2014 and SVM in Windows. The network was optimized with varied learning rate and different hidden layer nodes as suggested. The ANN (6:100:5) excludes NN50, pNN50, and mHR out of nine parameters. The input data were divided into 80% training, 10% validation, and 10% testing sets. The flowchart for graphical user interface (GUI) was shown [Figure 1]. A GUI was designed using MATLAB 2014 to evaluate the PS scale as per the clinicians' input values of the most promising parameters of time-domain HRV (SDNN and RMSSD), and its layout was presented [Figure 2].
One-way analysis of variance (ANOVA) and student t-test were performed using R.3.0.1 (R Foundation for Statistical Computing, Vienna, Austria) at the significance of P < 0.05. All the HRV indices were shown as mean ± standard error. Few measures were also tested and found significant at 0.0001 and were termed as highly significant.
| » Results and Discussion|| |
The obtained results suggested that, with increase in disease severity, the value of mRR interval, SDNN, RMSSD, TI, TiNN, SDHR, NN50, and pNN50 decreases. However, the increase in mean HR was observed from control to ECOG4 stage of LC, as shown in [Figure 3]. Although exception was also observed in the control group having lower values than ECOG3 and ECOG2 in TiNN and SDHR, respectively. Further, control and ECOG1 values were lower than ECOG2 in NN50 and pNN50. In line with the observed results, published literature revealed low HRV with decrease in SDNN and RMSSD values with cancer.,, Apart from SDNN and RMSSD, other HRV indices also reflect the alteration in ANS tone with vagal modulation and can be effective in suggesting cancer severity. A previous review also suggested reduced HRV levels with lower PNS activity among different types and progression of cancer stages., The neuronal vagal nerve controls the activity of PNS under cancer-mediated stress through the release of norepinephrine, dopamine, and bradykinin that triggers or inhibits the vascular endothelial growth factor (VEGF). The release of these neurotransmitters also modulates vascular elasticity (dilation and constriction), resulting in increase or decrease in hemodynamic parameters such as blood pressure and HR. With cancer stage progression, norepinephrine level also increase, triggering VEGF expression and leading to angiogenesis. Several outputs of ECOG PS scales obtained from GUI with SDNN and RMSSD as input has been shown in [Figure 4]. Age did not have an impact in this study as we included participants from various age groups. The circadian rhythm and posture affects HRV analysis, but to maintain the uniformity of data analysis, the measurements are taken in the supine position and during the day from 11 a. m. to 1 noon. Furthermore, healthy controls exhibit a reverse value of HRV compared to LC participants, especially in ECOG4. It is observed that patients suffering from cardiovascular disorder, hypertension, and diabetes have reduced HRV compared to healthy controls. A limitation of our study is that there were less participants in ECOG1 and ECOG2. In future, more participants can be considered in this categories.
|Figure 3: Changes in (a) mean RR interval (b) mean standard deviation of normal-to-normal interval (c) mean heart rate (d) root mean square of successive differences (e) heart rate variability triangular index (f) triangular interpolation of NN intervals, (g) standard deviation of heart rate, (h) NN50, and (i) pNN50 with progression of lung cancer|
Click here to view
|Figure 4: Output obtained for heart rate variability time-domain measures from graphical user interface|
Click here to view
Statistically, the obtained features were also significant between the different stages of cancer (Student's t-test), as shown in [Table 1]. In ANOVA test, mRR (P = 1.31e-12, F = 19.58), SDNN (P = 2.91e-07, F = 10.288), Mean HR (P = 5.62e-11, F = 16.562), STD HR (P = 9.7e-04, F = 11.389), RMSSD (P = 4.75e-07, F = 9.9515), NN50 (P = 2.33e-10, F = 15.465), PNN50 (P = 5.36E-11, F = 16.15), TI (P = 2.2e-16, F = 30.169), and TINN (P = 2.2e-16, F = 33.57) were observed significant.
ANN was applied based on the observed HRV features to classify cancer patients. The architecture of 9:100:5 was chosen to include all input features. Whereas, in 6:100:5, three features (NN50, TI, and mHR) were neglected for poor individual accuracy in classification. The network was optimized with a varied learning rate (0.01–0.9) at fixed hidden layer nodes (100). The learning rate for which the highest accuracy was achieved was selected [Table 2]. The selected learning parameters were used to test the network at different hidden layer nodes. Overall, the percentage accuracy of 85.98% and 89.64% was obtained at learning rates of 0.8 and 0.04 for 9 and 6 input nodes with regression analysis, respectively. Further, 83.37% and 86.64% were tested at 0.8 and 0.9 learning rate for 9:100:5 and 6:100:5 architecture, respectively. Thus, the ANN (6:100:5) could classify LC patients into five different stages with an accuracy of 89.64% [Table 3]. Moreover, the output accuracy of 100% was observed with SVM technique. HRV-based automated classification of LC is the novel method presented in this study.
|Table 2: Effects of learning rate on effectiveness of 9:100:5 and 6:100:5 network with fixed number of hidden neurons at 100 (number of iterations=1000)|
Click here to view
|Table 3: Effects of number of hidden layer nodes on performance of artificial neural network (9:100:5 and 6:100:5) at learning rate of 0.8 and 0.04 for regression analysis and 0.8 and 0.9 for confusion matrix analysis, respectively (testing results)|
Click here to view
| » Conclusion|| |
LC causes lower HRV through autonomic dysfunction. The features of HRV reflect the change in ANS tone with disease severity. These indices may be used in the design of intelligent system in the early prognosis of LC. Thus, with such systems, mortality rate can be reduced to a significant extent. The need of an expert clinician cannot be neglected. Automated systems and GUI will aid oncologists in their diagnosis.
Authors are grateful to Dr. Rajesh Singh (Professor and Head, Indira Gandhi Institute of Medical Sciences, Cancer Centre, Patna, India), Dr. Seema, Dr. Richa Madhavi, and Dr. Dinesh Sinha (Assistant Professor). Furthermore, authors express their gratitude to medical oncologist Dr. Shreeniwas Raut in HMRI Paras Hospital, Patna, for their clinical inputs and permitting for data collection in the hospital. Authors are also thankful to Dr. Rakesh Kumar (Professor, Department of Bio-Engineering, Birla Institute of Technology, Mesra, Ranchi, India) for his technical inputs for the work.
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
| » References|| |
Kim DH, Kim JA, Choi YS, Kim SH, Lee JY, Kim YE. Heart rate variability and length of survival in hospice cancer patients. J Korean Med Sci 2010;25:1140-5.
Weinberg RA. E2F and cell proliferation: A world turned upside down. Cell 1996;85:457-9.
Garland LH, Beier RL, Coulson W, Heald JH, Stein RL. The apparent sites of origin of carcinomas of the lung. Radiology 1962;78:1-11.
Couck MD, Brummelen DV, Schallier D, Greve JD, Gidron Y. The relationship between vagal nerve activity and clinical outcomes in prostate and non-small cell lung cancer patients. Oncol Rep 2013;30:2435-41.
Stewart BW, Kleihues P, editors. World Cancer Report. Lyon: IARC Press; 2003.
Schlesinger RB, Lippmann M. Selective particle deposition and bronchogenic carcinoma. Environ Res 1978;15:424-31.
Shukla, RS, Aggarwal Y. Spectral analysis to evaluate the effect of treatment on autonomic nervous system in pulmonary metastasis, Int J Eng Technol Sci Res 2017;4:501-6.
Guo Y, Koshy S, Hui D, Palmer JL, Shin K, Bozkurt M, et al.
Prognostic value of heart rate variability in patients with cancer. J Clin Neurophysiol 2015;32:516-20.
Li S, Sun Y, Gao D. Role of the nervous system in cancer metastasis. Oncol Lett 2013;5:1101-11.
Walsh D, Nelson KA. Autonomic nervous system dysfunction in advanced cancer. Support Care Cancer 2002;10:523-8.
Aggarwal Y, Singh N, Ghosh S, Sinha RK. Eye gaze-induced mental stress alters the heart rate variability analysis. J Clin Eng 2014;39:79-89.
Acharya RU, Lim CM, Joseph P. Heart rate variability analysis using correlation dimension and detrended fluctuation analysis. ITBM RBM 2002;23:333-9.
Yeh RG, Chen GY, Shieh JS, Kuo CD. Parameters investigation of detrended fluctuation analysis for short-term human heart rate variability. J Med Biol Eng 2010;30:277-82.
Caro-Morán E, Fernández-Lao C, Galiano-Castillo N, Cantarero-Villanueva I, Arroyo-Morales M, Díaz-Rodríguez L. Heart rate variability in breast cancer survivors after the first year of treatments: A case-controlled study. Biol Res Nurs 2016;18:43-9.
Chiang JK, Kuo TB, Fu CH, Koo M. Predicting 7-day survival using heart rate variability in hospice patients with non-lung cancers. PLoS One 2013;8:e69482.
Schubert C, Lambertz M, Nelesen RA, Bardwell W, Choi JB, Dimsdale JE. Effects of stress on heart rate complexity – A comparison between short-term and chronic stress. Biol Psychol 2009;80:325-32.
Rajendra Acharya U, Paul Joseph K, Kannathal N, Lim CM, Suri JS. Heart rate variability: a review. Med Biol Eng Comput 2006;44:1031-51.
Al-Shamasneh AR, Obaidellah UH. Artificial intelligence techniques for cancer detection and classification: Review study. Eur Sci J 2017;13:342-70.
Prabhu SV, Shaikh JA. Artificial neural network-based classification system for lung nodules on computed tomography scans. Int J Eng Sci Comput 2017;7:5097-599.
Aggarwal Y, Singh N, Sinha RK. Electrooculogram based study to assess the effects of prolonged eye fixation on autonomic responses and its possible implication in man-machine interface. Health Technol 2012;2:89-94.
Tarvainen MP, Niskanen JP, Lipponen JA, Ranta-Aho PO, Karjalainen PA. Kubios HRV – Heart rate variability analysis software. Comput Methods Programs Biomed 2014;113:210-20.
Shukla, RS, Aggarwal Y. Heart rate variability time-domain analysis in pulmonary metastasis to assess performance status. Indian J Sci Res 2017;14:540-5.
Aggarwal Y, Karan BM, Das BN, Aggarwal T, Sinha RK. Backpropagation ANN-based prediction of exertional heat illness. J Med Syst 2007;31:547-50.
Hanselman D, Littlefield BC. Mastering MATLAB 5: A Comprehensive Tutorial and Reference. Upper Saddle River, NJ, USA: Prentice Hall; 1997.
Zhou X, Ma Z, Zhang L, Zhou S, Wang J, Wang B, et al.
Heart rate variability in the prediction of survival in patients with cancer: A systematic review and meta-analysis. J Psychosom Res 2016;89:20-5.
De Couck M, Gidron Y. Norms of vagal nerve activity, indexed by Heart Rate Variability, in cancer patients. Cancer Epidemiol 2013;37:737-41.
Cowan MJ, Pike K, Burr RL. Effects of gender and age on heart rate variability in healthy individuals and in persons after sudden cardiac arrest. J Electrocardiol 1994;27 Suppl (1):1-9.
Huikuri HV, Niemelä MJ, Ojala S, Rantala A, Ikäheimo MJ, Airaksinen KE. Circadian rhythms of frequency domain measures of heart rate variability in healthy subjects and patients with coronary artery disease. Effects of arousal and upright posture. Circulation 1994;90:121-6.
Musialik- ydka A, Sredniawa B, Pasyk S. Heart rate variability in heart failure. Kardiol Pol 2003;58:10-6.
[Figure 1], [Figure 2], [Figure 3], [Figure 4]
[Table 1], [Table 2], [Table 3]