Abstract
With the dramatic increase of the worldwide threat of dengue disease, it has been very crucial to correctly diagnose the dengue patients in order to decrease the disease severity. However, it has been a great challenge for the physicians to identify the level of risk in dengue patients due to overlapping of the medical classification criteria. Therefore, this study aims to construct a noninvasive diagnostic system to assist the physicians for classifying the risk in dengue patients. Systematic producers have been followed to develop the system. Firstly, the assessment of the significant predictors associated with the level of risk in dengue patients was carried out utilizing the statistical analyses technique. Secondly, Multilayer perceptron neural network models trained via Levenberg-Marquardt and Scaled Conjugate Gradient algorithms was employed for constructing the diagnostic system. Finally, precise tuning for the models’ parameters was conducted in order to achieve the optimal performance. As a result, 9 noninvasive predictors were found to be significantly associated with the level of risk in dengue patients. By employing those predictors, 75% prediction accuracy has been achieved for classifying the risk in dengue patients using Scaled Conjugate Gradient algorithm while 70.7% prediction accuracy were achieved by using Levenberg-Marquardt algorithm.
Similar content being viewed by others
References
Monath, T. P., Dengue: the risk to developed and developing countries. Proc. Natl. Acad. Sci. USA. 91:2395-2400, 1994.
Gubler, G. J., Epidemic dengue/dengue hemorrhagic fever as a public health, social and economic problem in the 21st century. Trends Microbiol. 10:100-103, 2002.
World Health Organization, Dengue haemorrhagic fever: diagnosis, treatment, prevention and control, 2nd ed., Geneva, 1997.
Wallace Hazel, G., et al., Dengue haemorthagic fever in Malaysia 1973 epidemic. South East Asian J. Trop. Med. Pub. Hlth. 11:1-12, 1980.
Ministry of Health Malaysia, Statement on dengue by director general of health, 2009.
Ng, C. F. S. et al., Clinicians’ diagnostic practice of dengue infections. J. Clin. Virol. 40(3):202–206, 2007.
Ibrahim, F., Prognosis of dengue fever and dengue hemorrhagic fever using bioelectrical impedance, PhD thesis, University of Malaya, Malaysia, 2005a.
Ibrahim, F., Taib, M. N, Wan Abas, W. A. B., Chan, C. G., and Sulaiman, S., A Novel approach to classify risk in Dengue Hemorrhagic Fever (DHF) using bioelctrical impedance, IEEE Transaction on instrumentation and measurement. 54(1), 2005b.
Ibrahim, F., Taib, M. N., Wan Abas, W. A. B., Chan, C. G., and Sulaiman, S., A Novel Dengue fever (DF) Dengue and Hemorrhagic fever (DHF) Analysis using artificial neural network. Comput. Meth. Programs Biomed. 79:273-281, 2005c.
Faisal, T., Taib, M. N., and Ibrahim, F., Reexamination of risk criteria in dengue patients using the self-organizing map. Med. Biol. Eng. Comput. 48:293-301, 2010.
Faisal, T., Ibrahim, F., and Taib, M. N., A noninvasive intelligent approach for predicting the risk in dengue patients. Expert Syst. Appl. 37:2175-2181, 2010.
Icer, S., Kara, S., and Guven, A., Comparison of multilayer perceptron training algorithms for portal venous doppler signals in the cirrhosis disease. Expert Syst. Appl. 31:406-413, 2006.
Engin, M., Demirag, S., Zeki Engin, E., Celebi, G., Ersan, F., Asena, E., and Colakoglu, Z., The classification of human tremor signals using artificial neural network. Expert Syst. Appl. 33:754-761, 2007.
Gil Mendez, D., Johnsson, M., Manuel Garcia Chamizo, J., Soriano Paya, A., and Ruiz Fernandez, D., Application of artificial neural networks in the diagnosis of urological dysfunctions. Expert Syst. Appl., 2008.
Qiu, X., Tao, N., Tan, Y., and Wu, X., Constructing of the risk classification model of cervical cancer by artificial neural network. Expert Syst. Appl. 32:1094-1099, 2007.
Hagan, M. T., and Menhaj, M. B., Training feedforward networks with the Marquardt algorithm. IEEE Trans. Neural Netw. 5(6):989-993, 1994.
Moller, A. F., A scaled conjugate gradient algorithm for fast supervised learning, Neural Netw. 6:525-533, 1993.
Ampazis, N., and Perantonis, S. J., Levenberg-Marquardt algorithm with adaptive momentum for the efficient training of feedforward Neural Networks, Proceedings of the IEEE-INNS-ENNS International Joint Conference, pp. 126–131, 2000.
I. MathWorks, Neural Network Toolbox. MATLAB 7.0, Release 14, 2004.
Haykin, S., Neural networks: a comprehensive foundation, 2nd Edition: Prentice Hall, New Jersey, 1999.
Charalambous, C., Conjugate gradient algorithm for efficient training of artificial neural networks, IEE Proceedings, no. 139, pp. 301–310. 1992.
Gill, P. E., and Murray, W., Safeguarded steplength algorithms for optimization using descent methods, NPL Report NAC 37: National Physical Laboratory, Division of Numerical Analysis and Computing, Middlesex, England, 1974.
Delen, D., Sharda, R., and Bessonov, M., Identifying significant predictors of injury severity in traffic accidents using a series of artificial neural networks. Accid. Anal. Prev. 38:434–444, 2006.
Kohavi, R., A study of cross-validation and bootstrap for accuracy estimation and model selection. The Fourteenth International Joint Conference on Artificial Intelligence (IJCAI). Montreal, Quebec, Canada, pp. 1137–1145, 1995.
Feng, C. X., Yu, Z. G., Kingi, U., and Baig, M. P., Threefold vs. fivefold cross validation in one-hidden-layer and two-hidden-layer predictive neural network modelling of machining surface roughness data. J. Manuf. Syst. 24(2):93-107, 2005.
Hanley, J. A., and Mcneil, B. J., The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology. no. 143, vol. 29, 1982.
Metz, C. E., Basic principles of ROC analysis. Semin. Nucl. Med. 8:283-298, 1978.
Maciej, A., Piotr, A., Jacek, M., Joseph, Y., Jay, A., and Georgia, D., Training neural network classifiers for medical decision making: The effects of imbalanced datasets on classification performance. Neural Netw. 21:427-436, 2008.
Oxford University Press. Concise medical dictionary, 3rd ed., Oxford, 1990.
Swets, J. A., Measuring the accuracy of diagnostic systems. Science. 240:1285–1293, 1988.
Acknowledgment
This work is financially supported by a Malaysian Ministry of Science Technology and Innovation (MOSTI) Science Fund Project No. 11-02-03-1014 and postgraduate research Fund (PPP) No. PS138-2008B, University of Malaya.
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
Simple logistic regression
-
➢ Null hypothesis states that there will be no relationship between the probability and the independent variable.
-
➢ The -2Loglikelihood vale for each variable was tested with a null model.
-
➢ The significant variable was selected according the chi-square value at significant level of 0.25
Null model
No | Variables | B | Significance(sig.) | -2 Loglikelihood |
1 | constant | −0.22 | 0.015 | 694.076 |
Significant at p = 0.25 (1DF: chi-square 1.323)
Variables | B | -2 Loglikelihood | G-value | |
Gander | −0.317 | 690.991 | 3.086 | |
Weight(w) | 0.026 | 672.584 | 21.493 | |
Phase Angle (PA) | −0.028 | 693.964 | 0.113 | * |
Body Capacitance (BC) | 0.001 | 689.308 | 4.768 | |
Resistance (RES) | −0.005 | 667.485 | 26.592 | |
Reactance (REACT) | −0.032 | 675.590 | 18.487 | |
Extracellular Mass (ECM) | −0.066 | 674.458 | 19.618 | |
Body Cell Mass (BCM) | −0.036 | 688.238 | 5.839 | |
Lean body mass (LBM) | −0.033 | 680.855 | 13.222 | |
Fat Mass (FM) | 0.033 | 680.855 | 13.222 | |
(ERB) = (ECM/BCM) | 1.560 | 684.763 | 9.313 | |
Body Mass Index (BMI) | 0.119 | 647.612 | 46.465 | |
Basal Metabolic Rate (BMR) | 0.001 | 687.165 | 6.912 | |
Total Body Water (TBW) | 0.022 | 688.420 | 5.657 | |
TRT = TBW/W | −0.037 | 683.804 | 10.272 | |
Extracellular Water (ECW) | 0.071 | 677.375 | 16.702 | |
Intracellular Water (ICW) | −0.071 | 677.341 | 16.735 | |
ERI = ECW/ICW | 2.329 | 671.340 | 22.736 | |
DAY | 690.512 | 3.565 | * | |
DAY(1) | 0.289 | p = 0.47 | ||
DAY(2) | 0.299 | |||
DAY(3) | 0.275 | |||
DAY(4) | 0.727 | |||
Headache | 0.577 | 688.237 | 5.840 | |
Dizziness and fainting (dizz/fain) | 0.350 | 692.043 | 2.033 | |
Weakness lower limb (wllimb) | 0.679 | 684.180 | 9.897 | |
Arthralgia | 0.836 | 685.191 | 8.886 | |
Myalgia | 0.865 | 683.291 | 10.785 | |
Body ache | 0.591 | 685.800 | 8.277 | |
Nausea | −0.108 | 693.888 | 0.189 | * |
Vomit | 0.293 | 692.199 | 1.878 | |
Anorexia | 0.332 | 689.065 | 5.012 | |
Abdominal Epigastic pain (gastric) | 0.782 | 676.656 | 17.421 | |
Petechiea Rash (p.rash) | −0.316 | 691.154 | 2.922 | |
Flush face (flushf) | 0.079 | 693.962 | 0.115 | * |
Bleeding tendency (bt) | 0.974 | 669.752 | 24.325 | |
Chill and rigor (chillnr) | 0.630 | 693.592 | 0.485 | * |
Hepatomegaly (hepa) | 0.377 | 691.557 | 2.520 | |
Macular | 0.153 | 692.233 | 1.844 |
Simple linear Correlation and Correlation Coefficient for continuous variable
Pearson’s correlation value (r) between the significant parameters (r ≥ ±0.8)
Linearity test
Estimated probability versus independent variables
Predicted Probability of the Risk versus the Resistance
Predicted Probability of the Risk versus the Reactance
Predicted Probability of the Risk versus the Body Cell Mass
Predicted Probability of the Risk versus the Extracellular Mass
Predicted Probability of the Risk versus the Body Mass Index
Predicted Probability of the Risk versus ratio of the Total Body Water and the weight
Predicted Probability of the Risk versus the ratio of Extracellular Water /Intracellular Water
BMI groups
Group | Category | Frequency |
0 | Less than 18.5 | 85 |
1 | 18.5–21 | 90 |
2 | 21.1–24.9 | 148 |
3 | 25–29.9 | 96 |
4 | Equal or more than 30 | 79 |
Results from simple logistic test for Categorized BMI
Variables | B | -2 Loglikelihood | G-value compared with the null model |
NBMI | 641.373 | 52.703 | |
NBMI(1) | 0.704 | ||
NBMI(2) | 0.684 | ||
NBMI(3) | 1.887 | ||
NBMI(4) | 1.754 |
Fraction polynomial method chi-square(1df = 3.841, 3df = 7.815)
Summary of the use of the Fractional Polynomial Method for Resistance
No | Variables | df | -2 loglikelihood | G-value for model versus linear |
1 | RES | 1 | 667.485 | |
2 | RES,RES3 | 2 | 665.278 | 2.206 |
3 | RES,RES3,RES−2 | 4 | 665.278 | 2.206 |
Summary of the use of the Fractional Polynomial Method for Reactance
No | Variables | df | -2 loglikelihood | G-value for model versus linear |
1 | REACT | 1 | 675.589 | |
2 | REACT,REACT3 | 2 | 672.406 | 3.182 |
3 | REACT,REACT3,REACT−2 | 4 | 672.398 | 3.191 |
Summary of the use of the Fractional Polynomial Method for Body Cell Mass
No | Variables | df | -2 loglikelihood | G-value for model versus linear |
1 | BCM | 1 | 674.458 | |
2 | BCM,BCM3 | 2 | 673.492 | 0.965 |
3 | BCM,BCM3,BCM−2 | 4 | 672.717 | 1.740 |
Summary of the use of the Fractional Polynomial Method for Extracellular Mass
No | Variables | df | -2 loglikelihood | G-value for model versus linear |
1 | ECM | 1 | 688.237 | |
2 | ECM,ECM3 | 2 | 686.969 | 1.268 |
3 | ECM,ECM3,ECM−2 | 4 | 686.080 | 2.157 |
Summary of the use of the Fractional Polynomial Method for Body Mass Index
No | Variables | df | -2 loglikelihood | G-value for model versus linear |
1 | BMI | 1 | 647.612 | |
2 | BMI,BMI3 | 2 | 646.654 | 0.957 |
3 | BMI,BMI3,BMI−2 | 4 | 646.601 | 1.010 |
Summary of the use of the Fractional Polynomial Method for the ratio of the Total Body Water and the Weight
No | Variables | df | -2 loglikelihood | G-value for model versus linear |
1 | TRT | 1 | 683.804 | |
2 | TRT,TRT3 | 2 | 682.241 | 1.562 |
3 | TRT,TRT3,TRT−2 | 4 | 681.791 | 2.013 |
Summary of the use of the Fractional Polynomial Method for the ratio of the Extracellular water and Intracellular Water (ICW)
No | Variables | df | -2 loglikelihood | G-value for model versus linear |
1 | ERI | 1 | 671.340 | |
2 | ERI,ERI3 | 2 | 666.254 | 5.085* |
3 | ERI,ERI3,ERI−2 | 4 | 665.269 | 6.071 |
ERI Groups
Group | Category | Frequency |
0 | 0.5–0.65 | 86 |
1 | 0.651–0.75 | 123 |
2 | 0.751–0.85 | 82 |
3 | 0.851–0.95 | 73 |
4 | Equal or more than 0.951 | 134 |
Results from simple logistic test for Categorized ERI
Variables | B | -2 Loglikelihood | G-value compared with the null model |
NERI | 674.267 | 19.810 | |
NERI(1) | 0.100 | ||
NERI(2) | 0.144 | ||
NERI(3) | 0.154 | ||
NERI(4) | 0.996 |
Multiple Logistic regression
Null hypothesis states that removing the variable from the model does not affect the model Summary of the removed variables
Variables | Removed Variables | df | -2 Loglikelihood | G-value compared with the full model | p-value |
model 1 | TRT, vomit | 2 | 561.468 | 0.592 | >0.05 |
model 2 | hepa,ECM,arthalgia | 3 | 562.076 | 0.608 | >0.05 |
model 3 | headache,macular, backache | 2 | 563.944 | 1.868 | >0.05 |
model 4 | anorexia,sex | 2 | 566.330 | 2.385 | >0.05 |
model 5 | myalgia | 1 | 568.799 | 2.470 | >0.05 |
model 6 | dizzfain | 1 | 570.180 | 1.381 | >0.05 |
Rights and permissions
About this article
Cite this article
Faisal, T., Taib, M.N. & Ibrahim, F. Neural network diagnostic system for dengue patients risk classification. J Med Syst 36, 661–676 (2012). https://doi.org/10.1007/s10916-010-9532-x
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10916-010-9532-x