A hybrid model based on modular neural networks and fuzzy systems for classification of blood pressure and hypertension risk diagnosis

Highlights

• A hybrid model using modular neural networks and fuzzy logic was designed.

• Risk of hypertension diagnosis of a person is obtained with the model.

• Hybrid systems for this kind of complex problems have excellent performance.

• Fuzzy logic helps model uncertainty in decision process of diagnosis.

Abstract

In this paper, a hybrid model using modular neural networks and fuzzy logic was designed to provide the hypertension risk diagnosis of a person. This model considers age, risk factors and behavior of the blood pressure in a period of 24 h, using as a basis the Framingham Heart Study.

Records of blood pressure are collected with the ambulatory blood pressure monitoring (ABPM), a device which takes readings for a period of time of 24 h. A modular neural network was designed, with three modules, of which the first and second modules correspond to the systolic and diastolic pressures and the last one to the heart rate. Each module is trained with the data obtained by the ABPM of different patients, this in order that the neural network learns the different behaviors that the blood pressure may have. Also, different architectures and learning methods are considered to obtain the best possible architecture. In addition, two fuzzy inference systems (FISs) for classification purpose are proposed, the first one for the heart rate level and the second one for the night profile of the patient. These were tested with different types of membership functions and then selecting the FIS that obtained the best results. Furthermore, a third FIS as a blood pressure classifier is also used.

The different proposed methodologies were tested, in the case of the modular neural network to find the architecture that produces better results and in the fuzzy inference systems to find which membership functions were the ideal ones for the case study, in this way obtaining overall good results. For the case of the modular neural network, the learning accuracy in the first module is 98%, in the second module is 97.62% and the third module is 97.83% respectively. For the night profile, the fuzzy system is compared to a traditional system of production rules, and it is noted that the first one gives all correct outputs and the second one just gives 53% of the outputs, this is due to the uncertainty handling that fuzzy systems can provide, which the traditional system cannot because its rules are very strict.

Hybrid intelligent systems for the solution of this kind of complex problems have excellent performance, due to the good learning in each module of the neural network and the classification uncertainty that is well managed by the fuzzy systems, obtaining with this a hybrid combination for achieving good results.

Introduction

Blood pressure (BP) is the force exerted against the walls of the arteries as the heart pumps blood, which is necessary for the blood to circulate through the blood vessels and provide the oxygen and nutrients necessary to all organs in order that the body can function properly (Rosendorff, 2013).

BP actually has two components: Systolic Pressure, which corresponds to the maximum value, measuring the force of the blood in the arteries when the heart contracts (beats), Diastolic pressure, which corresponds to the minimum value, measuring the strength of blood in the arteries while the heart is relaxed (filling with blood between the heartbeats) (Rosendorff, 2013).

In adults, normal blood pressure is defined as a systolic pressure below 139 mmHg and a diastolic pressure below 89 mmHg. It is normal for blood pressure to change, being lower at night with sleep and higher in the early hours of the morning (Rosendorff, 2013).

The Heart Rate (HR) is the number of times the heart contracts per minute. Normal heart rate undergoes healthy variation, due to certain conditions, such as exercise, body temperature, body position and emotions (“American Heart Association,” 2015) (MacGill, 2017)

High blood pressure or hypertension, is the elevation of blood pressure above the normal considered values, which means a systolic blood pressure (SBP) above 140 mmHg or a diastolic blood pressure (DBP) above 90 mmHg (Battegay, Lip, & Bakris, 2005).

Hypertension can be: Essential hypertension, which arises with no specific identifiable cause, but with a hereditary history, it may appear in isolation or be part of a complex of alterations that are found in insulin resistance (Beevers, Lip, & O’ Brien, 2007) (Carretero & Oparil, 2000). Secondary hypertension: this is a type of hypertension that has demonstrable causes, the most frequent causes of this type of hypertension are renovascular disease, endocrine, pregnancy, acute stress, renal failure and coarctation of the aorta, which can produce hypertension, and must be suspected in two situations: Hypertension in young people (< 35 years) and the absence of family history of hypertension (Beevers et al., 2007) (White, 2007).

For the blood pressure levels, the European guidelines for hypertension (Mancia et al., 2013) are used for classification. This is because these guidelines have not been widely used for this type of research, in addition to having more levels of classification in comparison with the American guidelines for hypertension management. The different classes of blood pressure are presented in Table 1, in accordance to the European guidelines for hypertension.

The Ambulatory Blood Pressure Monitoring (ABPM) is a non-invasive method for obtaining 24-h blood pressure measurement, which consists of a device connected to a sphygmomanometer cuff, which records the blood pressure of patients in a time interval, typically programmed from 15 to 20 min in the day and every 30 min during the night, and when the period of the reading ends, the information is downloaded into a computer (Guido, 2008) (Grossman, 2013).

The normal circadian profile is characterized by the decrease of between 10 and 20% of the nocturnal BP values compared to the daytime or activity values (dipper profile) (Friedman & Logan, 2009). The absence of a decrease in nocturnal BP < 10% is considered to be a non-dipper pattern. Another way of defining the dipper/non-dipper pattern is by using the night/day ratio, so that dipper patients would have a quotient between 0.90 and 0.80, non-dippers between 0.91 and 1.00, the dipper Extreme (nocturnal BP decrease > 20% of daytime BP) is 〈0.80 and the riser (mean of nocturnal BP values above the mean of daytime) has a quotient〉 1.00 (Feria-Carot & Sobrino, 2011).

In order to provide a patient's risk of developing hypertension, we based our work on the Framingham heart study (“Framingham Heart Study,” 2016) for the data. This study began in 1948 and was applied to a group of 5209 men and women from Framingham, Massachusetts ranging from 30 and 62 years old, who not had any cardiovascular disease or had not suffered a heart attack or cardiovascular accident. As time has passed the descendants of the original group have been added, and the objective of the Framingham Heart Study is the identification of the risk factors that influence the development of cardiovascular diseases (Kannel, 2000).

This study is based on a Weibull regression model, which is used to calculate risks in a given time, and for this case, the diagnosis of the risk of developing hypertension in a period of 4 years is given. In this study the variables such as age, sex, body mass index, if the patient has hypertensive parents smoking habit, systolic and diastolic blood pressure are taken into account (Kawada & Otsuka, 2010).

The risk of developing hypertension is given by the following expression:

$$FHSpredictorrisk=1-exp\left[-exp\left(\frac{\ln \left(4\right)-22.94954+\Sigma \mathrm{X}\beta }{0.8769}\right)\right]$$

Where:

• β: It is the coefficient of regression.

• X: It is the level for each variable.

If the gender is female then the variable is assigned a 1, and if it is male is assigned a 0.

If none of the parents is hypertensive is assigned a 0, if one of the parents is hypertensive is assigned 1 and if both are then is assigned a 2 (“Framingham Heart Study,” 2016).

At present, hybrid systems are a powerful tool for solving complex problems, because two or more soft computing techniques can be used simultaneously to solve a particular problem, and thereby reduce its complexity. In addition, hybrid systems are aimed at improving the efficiency and power of reasoning as well as the expressivity of isolated intelligent systems (Fdez Riverola & Corchado, 2003);(Medsker, 1995).

Soft computing is the opposite of hard computing, in that it is tolerant to imprecision, uncertainty, partial truth, and approximation. On the other hand, “nature-inspired” methodologies are designed to emulate one or more aspects of biological systems, and can be utilized for complementing both hard or soft computing.

For the model to be developed, soft computing techniques, such as artificial neural networks and fuzzy logic are used, since these methodologies have several characteristics in common, including the fact that they are free model estimators that can be adjusted or trained to improve their performance (Medsker, 1995).

An artificial neural network is an information processing system, which has certain performance characteristics in common with biological neurons. Artificial neural networks have been developed as generalizations of mathematical models of human cognition or neural biology (Samarasinghe, 2007).

In general, the implementation of modular neural networks is based on the "divide and conquer" principle, which consists of the decomposition of a task into less complex and smaller subtasks, so that every task is learned by different experts and then reuse the learning of every subtask to solve the entire problem (Melin & Castillo, 2005).

In order to perform a combination or integration of module results, modular neural networks use response integrators, such as the average, the winner takes all, fuzzy logic, gating network, among others (Jang, Sun, & Mizutani, 1997).

Moreover, in this research, fuzzy systems are used as response integrators, because in this way we can handle the uncertainty in the decision, of which a brief explanation is given below:

Fuzzy logic is a type of logic that involves approximate rather than exact modes of reasoning, and it can be viewed as an attempt to construct a model of human reasoning that reflects its approximate and qualitative character. Its ultimate goal is to provide the basis for approximate reasoning using imprecise propositions based on fuzzy set theory, similar to classical reasoning using precise propositions based on classical set theory (Zadeh, 1989).

The term fuzzy logic was introduced in 1965 with the publication of "Fuzzy Sets" (Zadeh, 1965) and proposed by Zadeh at the University of California at Berkeley for the journal Information and Control, which was based on the original work of J. Lukasiewicz (Chen & Pham, 2001) on multi-valued logic. Fuzzy logic aims at creating mathematical approximations for solving certain types of problems. This type of logic produces precise results from imprecise data, making it particularly useful in electronic or computational applications. The fuzzy adjective applied to this logic is due to the fact that the "non-deterministic" truth values used in this usually have a connotation of uncertainty.

Different researchers have implemented neural networks or fuzzy systems for predicting hypertension and other heart diseases (Ture, Kurt, Turhan Kurum, & Ozdamar, 2005) (Kurt, Ture, & Kurum, 2008) (Das & Ghosh, 2013) (Samant & Rao, 2013) (Salazar Mendiola, Vargas Luna, González Guerra, & Cortés Ramírez, 2013) (Seera & Lim, 2014) (Choudhury & Baruah, 2015) (Fraz et al., 2015) (McRae et al., 2016), and some are explained below:

For predicting hypertension, Assaghir, Janbain, Makki, Kurdi, and Karam (2017) have used a method based on neural networks, which considers ten risk factors as inputs, including gender, hear rate, body mass index, body fat, waist, hip, physical activities, smoke, salt and stress. On the other hand as outputs, the systolic and diastolic blood pressures are used. The results achieve more than 85% of prediction accuracy, which is acceptable in the diagnosis of systolic and diastolic blood pressure; this means that the complexity of the neural networks allowed the model to capture the relation between the risk factors and their effect on the blood pressure.

Mukherjee and Halder (2017) have presented a novel methodology using a fuzzy logic controller for indicating the appropriate administration of drug for the treatment of hypertensive and depressive persons. This controller responds very fast, so that the cardiac output does not decrease. The Mamdani fuzzy controller acts as endioprine hormone control and this means that depending on the range of hypertension in systolic and diastolic pressure is the percentage of drug that is administered.

For predicting hypertension without measurements, Wang, An, Chen, Li, and Alterovitz (2015) constructed a prediction model based on the hybrid use of logistic regression and artificial neural networks. For this, the binary logistic regression was used for selecting important risk factors to develop hypertension. Afterward, eleven risk factors are the inputs of the neural network and as output is the classification of hypertensive or normal patient. After the experimentation they can observe that neural network prediction model obtains over 72% accuracy, this indicates that the model has better stability and robustness than the logistic regression model.

Abrishami and Tabatabaee (2015) have proposed a fuzzy expert system and a multi-layer neural network system for diagnosis of Hypertension. In the fuzzy expert system, they used as input the systolic blood pressure and body mass index and as output the hypertension level of the patient. For the neural network they used five inputs, which are systolic blood pressure, smoking, age, weight and body mass index and the output is employed for the diagnosis of the hypertension, and finally the results of two systems are compared individually, obtaining good results for both expert systems.

For measuring health parameters of patients, Patil and Mohsin (2013) have proposed a wireless sensor network system for continuous monitoring of pulse and temperature of patients remotely or in the hospital, and it transmits the bio-signals to the Doctor and Patient mobile phone. Data stored in a database is send to the fuzzy logic controller (FLC) to improve accuracy and amount of data to be sent to the remote user. The FLC system receives context information from the sensor as input (the patient age and pulse), and output is the status of the patient pulse.

The main advantage of these previous works, focusing on fuzzy systems, is that granularity is simple, observing that the risk of hypertension is classified only into low, moderate or high. On the contrary the fuzzy system designed in the present paper has higher granularity, and this is because it is based on the European Hypertension Guidelines.

In the same way in these previous works, all the information is being worked on separately, observing that in the above mentioned works different factors are used, but all together with the artificial neural network or with the fuzzy systems.

The main contribution of this work is the proposed hybrid model combining modular neural networks with fuzzy systems for achieving accurate and efficient blood pressure classification and hypertension diagnosis. Previous models have only consider either fuzzy logic or neural networks, but not modular networks. In addition, even in previous hybrid neuro-fuzzy models, the hybridization is not with modular neural networks, which is an advantage of the proposed model that makes it faster for real time diagnosis and providing better accuracy. The use of fuzzy logic is to handle the uncertainty in the process of diagnosis, which is an inherent characteristic of this problem, which also helps in improving results. The models are built in such a way as to improve on the Framingham Heart score results, so that the risk of hypertension diagnosis for a patient can be provided in a four year window in advance with more accuracy.