Correlation between drinking water fluoride and TSH hormone by ANNs and ANFIS.

BACKGROUND: Artificial neural networks (ANNs) and adaptive neural-fuzzy Inference system (ANFIS) are the best solutions to finding the correlation between some water parameters and human hormones. The correlation between thyroid stimulating hormone (TSH) and drinking water fluoride studied by ANNS and ANFIS models in Yazd city.
METHOD: In this study, eighty people with thyroid gland disorder and 213 healthy people invited. Their thyroid hormones and fluoride drinking water analyzed.
RESULTS: The result of ANFIS showed R2 = 0.81 for test and R2 = 0.85 for train in all cases and controls data. This results were R2 = 0.73 and R2 = 0.81 for ANNs respectively.
CONCLUSION: This models can be used as an alternative for show correlation between Drinking Water Fluoride and TSH Hormone and R2 = 0.85 gained from ANFIS was the best.

Background

Fluoride, an anion of fluorine (halogen) is found in water. It has beneficial effects on teeth and bones at low concentrations but excessive exposure to fluoride has adverse effects [1-10]. There are ways to get fluoride, but water is the most important [11-14]. There are methods to remove fluoride from water that have been reported by researchers [15-18]. Peckham (2014a; 2015b) stated that fluoride intake of more than 0.01 mg/kg body weight/day impacts thyroid gland function because it has adverse effects on iodine. Since children drink less water than adults, 0.8 mg/L is recommended by the European Union Scientific Committee on Health and Environmental Risk [5, 6]. Iran’s recommended level of fluoride in drinking water is 0.5–1.5 8 mg/L based on WHO Guidelines for drinking-water quality [19]. Dey and Giri (2016) found that excessive exposure to fluoride caused increase thyroid-stimulating hormone (TSH) and increase T3 and T4 hormones especially in people who do not consume enough iodine, therefore causing hypothyroidism [2]. They illustrated that consuming 20 mg/L sodium fluoride per day by rats repressed the production of nucleic acids and Triiodothyronine (T3), thyroxin (T4) hormones and change the activities of some metabolic enzyme such as: K (+) - Na (+) -ATPase, thyroid peroxidase, and 5, 5′-deiodinase [2]. Rastogi and Monika (2014) said that T3 and T4 are two hormones significant for the production of proteins to adjust body temperature which generates total energy, regulate the thyroid gland and controls human body metabolism [20]. Researchers have reported that thyroid dysfunction is a low thyroid production (hypothyroidism) or excessive production of thyroid hormone (hyperthyroidism) [21-26]. Diagnosis of thyroid disease carried out by blood tests include the TSH, T4 test [22, 23, 27]. Governments are responsible for researching on the health challenges of its people due to reduced or excessive fluoride exposure [2]. Nowadays, computer technology and information have been developed in the sciences [28]. Therefore, artificial neural networks (ANNs) and adaptive neural-fuzzy Inference system (ANFIS) are the best solutions to finding the correlation between some water parameters and human hormones. ANNs are used for recognition, prediction, optimization, associative memory, and control [28]. ANNs have many advantages such as, flexible modeling structure for large data sets and highly accurate results that support clinical decision-making [29-31]. ANNs model was applied by some researchers in the diagnosis of thyroid disease [20–22, 24, 28, 48, and, 32, 33] and other diseases including diagnosis of tuberculosis [34], schizophrenia [36], anemia in children [37], heart disease [38] and, predicting prostate cancer [35]. Also, ANNS have been used by Lahner et al. (2008) in the recognition of the presence of thyroid disease in patients with atrophic body gastritis [39]. The aim of this study was to apply ANNS and ANFIS in modeling thyroid stimulating hormone (TSH) versus drinking water fluoride in Yazd city.

Methods

Study area, sample size, hormones analyze

The main source of drinking water in Yazd city is surface water and well resources in different seasons; thus, differences in fluoride concentration are bound to exist. Yazd city is the driest residential area in Iran [40]. Kheradpisheh et al. (2016) found that the mean ± SD of drinking water fluoride is 0.5 mg/L ± 0.27 in this area [3]. This nested case and control study selected cases and controls from Yazd Healthy Study (YaHS) which is a prospective cohort study. YaHS was conducted on ten thousand people aged 20 to 70 years in 2014 _ 2015. Details of Yazd Health Study has been published elsewhere [49]. The cluster sampling method was used in selecting cases with thyroid disease and medicine was not utilized. Control samples were participants with no thyroid disease aged 20 _ 60 years old. Eighty cases and 213 controls were selected. Informed consent was sought and obtained from the subjects upon which thyroid hormones test was conducted. Thyroid hormone analysis was carried out in Yazd Central Laboratory.

Drinking water fluoride study

Drinking water fluoride concentration was determined by SPADANS method number 8029 as described by Standard Methods for the Examination of Water and Wastewater [41]. All devices and reagents used in this study were obtained from HACH, Germany.

Application of ANNS and ANFIS model

The Neural Network Toolbox of Matlab© R2011a was used to create, train, and test data from ANNS and ANFIS model. Data were normalized. Inputs contained 268 out of every three T3, T4, and F from total cases and controls. Outputs were 268 out of TSH. Train GD, train GDA, train GDX, and train SCG, train LM were used as network training function, and R2, RMSE, NRMSE were obtained from test plot model and Test Scatter model for TSH hormone and also to train all cases and control data by ANNS and ANFIS model. The abbreviations are explained in Abbreviations list in this text. The training was achieved by a back propagation method after setting network properties and raining parameters. Back propagation method has been used on some research as Aktan et al. (2016); Boyacioglu et al. (2010); Rezaei Kahkha et al. (2016); Razia et al. (2016); Svalina et al. (2013) [22, 42–45]. The network was tested with thyroid hormones data collected from cases and controls and measured drinking water fluoride concentration.

Data Availability

Please contact the author for data requests.

Results

The main goal of this paper was to apply ANNS and ANFIS Models for TSH Hormone modeling versus Drinking Water Fluoride in Yazd city. In this study, R square, RMSE, and NRMSE were calculated for observed and predicted data in ANNS and ANFIS models. Figures 1 and 2 indicate the correlation diagram for Test Plot model and Test Scatter model for TSH hormone in all cases and control data by ANFIS Model. The results were: R2 = 0.81, RMSE = 0.6, NRMSE = 6.4. Figures 3 and 4 show the train plot model and train scatter model results for TSH hormone in all cases and controls data by ANFIS Model. The result were: R2 = 0.85, RMSE = 4.8, NRMSE = 9.3. The back propagation learning algorithm was used to minimize system error between real output and target output. The number of hidden layer are achieved by trial and error method in the ANNS and ANFIS Models. Figures 5 and 6 show the test plot model and test scatter model results for TSH hormone in all cases and control data by ANNS Model. The results were: R2 = 0.73, RMSE = 0.8, NRMSE = 8.4. Figures 7 and 8 show the train plot model and results for TSH hormone in all cases and controls data by ANNS Model. The results were: R2 = 0.81, RMSE = 5.15, NRMSE = 9.5.The number of Epoch was1000, the number of hidden layer was 9, the number of train algorithm was 5 contain: train GD, train GDA, train GDX, train SCG, and train LM for ANNS and ANFIS models. The best train algorithm was train LM. 268 out of 293 (80 cases and 213 controls) cases were used in the ANNS and ANFIS models as the age was not within our desirable range or the results of all three hormones were not acceptable. Inputs were 268 out of every three T3, T4, and F from total cases and controls. Outputs were 268 out of TSH. Test Inputs were 79 out of every three T3, T4, and F from total cases and control. Test Outputs were 79 out of TSH. Train Inputs were 189 out of every three T3, T4, and F from total cases and control. Outputs were 189 out of TSH.

Fig. 1

Test Plot model for TSH hormone in all data (Cases and controls) by ANFIS. R2 = 0.81, RMSE = 0.6, NRMSE = 6.4

Fig. 2

Test Scatter model for TSH hormone in all data (Cases and controls) by ANFIS.R2 = 0.81, RMSE = 0.6, NRMSE = 6.4

Fig. 3

Train Plot model for TSH hormone in all data (Cases and controls) by ANFIS. R2 = 0.85, RMSE = 4.8, NRMSE = 9.3

Fig. 4

Train Scatter model for TSH hormone in all data (Cases and controls) by ANFIS. R2 = 0.85, RMSE = 4.8, NRMSE = 9.3

Fig. 5

Test Plot model for TSH hormone in all data (Cases and controls) by ANNS. R2 = 0.73, RMSE = 0.8, NRMSE = 8.4

Fig. 6

Test Scatter model for TSH hormone in all data (Cases and controls) by ANNS. R2 = 0.73, RMSE = 0.8, NRMSE = 8.4

Fig. 7

Train Plot model for TSH hormone in all data (Cases and controls) by ANNS. R2 = 0.81, RMSE = 5.15, NRMSE = 9.5

Fig. 8

Train Scatter model for TSH hormone in all data (Cases and controls) by ANNS. R2 = 0.81, RMSE = 5.15, NRMSE = 9.5

Discussion

Results show that outcomes of ANNs and ANFIS models are in good agreement with experimental data; this indicates that ANNs and ANFIS models can be used as an alternative for the determination of correlation between TSH hormone and drinking water fluoride. R2 = 0.85 was the best value obtained from the ANFIS model. In this paper, by selecting 9 neurons in the hidden layer, we can reach 81% classification accuracy for thyroid disease by ANNs and 85% by ANFIS. This paper presents ANFIS model as an assistance system for the determination of correlation between TSH and fluoride. This model used drinking water fluoride, an environmental factor, as an input data and TSH hormone value as output of the model which determine hypothyroidism like an endocrinologist expert. To determine the correlation between TSH hormone and how other environmental factors play an important role in the prevention of hypothyroidism especially in places with high fluoride in the water. In spite of the fact that, many successful studies using neural networks and ANFIS for classification including medical diagnosis like: (Adeli and Neshat (2010); Caturegli et al. (2014); Emin Aktan et al. (2016); Omitek et al., 2013); Prerana et al. (2015); Rastogi and Monika (2014); Saylam et al. (2013); Soleimanian et al. (2013) [20–22, 24, 28, 33, 46, 48], But in this paper, we presented the correlation between disease and drinking water factor that we think is more important especially in the study area. The results clearly show that neural networks as traditional statistical methods can estimate these correlations.

As Galletti and Joyet (1958) pointed out, fluoride was used to treat hyperthyroidism in South America and Europe in 19 decades. Iodine is used by the thyroid gland to make T3 and T4 hormones. When iodine is low in the body, other available halogens such as bromine, chlorine, and fluorine will be absorbed into and block the thyroid gland [47]. In areas that experience excess fluoride, especially from water, low iodine levels in the body can cause fluoride uptake into the thyroid gland. Thus, to help our thyroid function, we must consider limiting fluoride exposure and adding iodine to our diet [47]. As mentioned by Basha et al. (2011) and Dey and Giri (2016), fluoride can block the iodine receptors in the thyroid gland therefore, we are suggest a study can be done on using pure treated drinking water by hypothyroidism when they are treated by medication. The thyroid gland is the most sensitive tissue in the body to fluoride [2]. Basha et al. (2011) and Dey and Giri (2016) stated that TSH hormone as well as T3 and T4 hormones decreased due to excess fluoride concentration and thus cause hypothyroidism in some populations [1, 2]. This paper showed this reality that the value of TSH hormone increased by increasing water fluoride concentration. Our results are in line with those of other researches such as Basha et al. (2011) and Dey and Giri (2016).

Conclusion

Essentially, it means that the ANNS and ANFIS methods can be used to modeling thyroid stimulating hormone (TSH) versus drinking water fluoride. The proposed method in this paper can be a solution to increase the performance of ANNS and ANFIS. Therefore, it can be generalized to determine other factors that affect TSH hormone by models. For further studies and investigation, we propose using sex, age, physical conditions, environmental factors especially drinking water etc., as effective factors in thyroid diagnosis.