A Novel Heart Disease Prediction System Based on Quantum Neural Network Using Clinical Parameters

the study the of the of the SS and AKG the analyses of the study. ABSTRACT Aims: The diagnosis of Heart disease at earliest possible stage is very crucial to increase the chance of successful treatment and to reduce the mortality rate. The interpretation of cardiovascular disease is time-consuming and requires analysis by an expert physician. Thus there is a need of expert system which may provide quick and accurate prediction of Results: The obtained result of Heart disease prediction match with the expert physician’s opinion with 96.97% accuracy and shows high degrees of sensitivity and specificity. Conclusion: The proposed Heart Disease Prediction System based on Quantum Neural Network gives the high degrees of accuracy in predicting the risk of cardiovascular diseases, are also the best results based on clinical factors. The result generated by this system has been evaluated and validated on data of patients with the Doctor’s diagnosis. This system will help the doctors to plan for a better medication and provide the patient with early diagnosis as it performs reasonably well even without retraining. Such an expert system may also prove useful in combination with other systems to providing diagnostic and predictive medical opinions in a timely manner.


INTRODUCTION
Heart disease is one of the major causes of death worldwide. According to World Health Organization's report, cardiovascular diseases (CVD) are the number one cause of death globally. More people die annually from CVDs than from any other cause. An estimated 17.3 million people died from CVDs in 2008, representing 30% of all global deaths. Low and middle-income countries are disproportionally affected: over 80% of CVD deaths take place in low and middle-income countries and occur almost equally in men and women [1].
Nowadays, many hospitals use some sort of information systems to manage their healthcare or patient data [2]. Unfortunately, these data are infrequently used to support clinical decision making. The main objective is to develop an Intelligent Heart Disease Prediction System with Quantum Neural Network using heart disease databases to make intelligent clinical decisions which cannot be done in traditional decision support systems.
In this study, we used Quantum Neural Network for the identification of CVD patients and it also predicts the rate of risk. The conventional Neural Network differs from the Quantum Neural Network (QNN) in the form of nonlinear activation functions of their hidden units. A multilevel activation function is used in QNN instead of the ordinary sigmoid function. Each multilevel function consists of the sum of sigmoid functions shifted by the quantum intervals. All models were trained and tested on the test dataset using clinical risk factors.
The major clinical risk factors that are used in our model can be modified, treated or controlled by changing our lifestyle or taking medicine and their normal reference ranges are shown in Table 1.

Diabetes Mellitus (DM)
Diabetes seriously increases the risk of developing cardiovascular disease. Both major types of diabetes mellitus, Type 1 (insulin dependent) diabetes and Type 2 (non-insulindependent) diabetes are associated with a significantly increased risk of coronary heart disease, cerebrovascular disease and peripheral vascular disease [3][4][5][6][7][8]. Even when glucose (blood sugar) levels are under control, diabetes increases the risk of heart disease and stroke, but the risks are even greater if blood sugar is not well controlled. About three -quarters of people of world population are diabetic. In this situation it is extremely important to work with healthcare provider to manage it. Coronary Artery Disease is a major complication of both insulin -dependent diabetes mellitus (IDDM) and non-insulin -dependent diabetes mellitus (NIDDM). In 14 years of follow-up, the Rancho Bernardo Study, in which 334 men and women with NIDDM were compared with 2137 men and women without diabetes, the relative risk for Coronary Artery Disease deaths was 1.9 in diabetic men and 3.3 in diabetic women compared with non-diabetic men and women after adjustment for other Coronary Artery Disease risk factors [9]. The relation between diabetes and cardiovascular diseases are not uniform in all populations. In the WHO Multinational Study of vascular Disease in Diabetics, the incidence of death in diabetic patients that was attributable to circulatory disease ranged from 32 percent in men and 0 percent in women in Tokyo to 67 percent in men and 47 percent in women in London [10].

Low Density Lipoprotein (LDL)
Low density lipoprotein (LDL) is one of the most fascinating macromolecular assemblies ever evolved by nature and has consequently attracted the attention of scientists for decades. As the principal transporter for cholesterol in the blood, circulating LDL guarantees a constant supply of cholesterol for tissues and cells and is an essential constituent of human life. The cellular uptake of LDL is mediated by specific receptors, either classical B/E receptor or the scavenger receptor pathway [11,12].

High Density Lipoprotein (HDL)
Abnormality of high density lipoprotein may increase the risk of atherosclerosis. Studies have shown that high-density lipoprotein (HDL) cholesterol levels are a strong, independent inverse predictor of cardiovascular disease [13][14][15][16][17]. According to the Framingham Heart Study, HDL level was considered as more persuasive risk factor for coronary heart disease than the level of low-density lipoprotein (LDL) [16].

Triglycerides (TG)
High triglycerides raise the risk of heart disease. The American Heart Association (AHA) recommends that a triglyceride level of 100 mg/dL(1.3 mmol/L) or lower is considered "optimal." The AHA says this optimal level would improve heart health. However, the AHA doesn't recommend drug treatment to reach this level. Instead, for those trying to lower their triglycerides to this level, lifestyle changes such as diet, weight loss and physical activity are encouraged. That's because triglycerides usually respond well to dietary and lifestyle changes.

Related Work
Many risk assessments are based on relatively simplistic strategies, the clinician may identify whether a "risk factor" is present. However, national guidelines advocate the use of scoring systems for CHD risk. Sophisticated methods like the Framingham Risk Score (FRS), the European Systematic Coronary Risk Evaluation (SCORE), the Sheffield risk and treatment table allow calculation of an individual's risk, as a function of the patient's values on selected established coronary heart disease risk factors [18]. The variables used for risk score computation by the different models used, are listed below:
The Framingham equations used were developed from cohort data collected in the 1960s and 1970s. We know that since then there has been a fall in the incidence of CHD mortality and a change in the prevalence of smoking. Whilst it may be that the change in incidence is due to the change in the prevalence of risk factors included in the Framingham equation, we cannot be sure that there are not other extraneous factors that have varied over that time and may have affected the incidence of CHD. If so, the validity of the Framingham equation in modern populations may be undermined. For example, some other risk factors such as serum fibrinogen levels, homocysteine, chlamydia infection or any undetected factors may have altered which would distort the Framingham predictions.

The Joint British Cardiac Society (Bcs)/ British Hypertension Society (Bhs)/ British Hyperlipidemia Association (Bha) Risk Score
The variables used for risk score computation are Gender, age, Total Cholesterol, HDL, Systolic Blood Pressure (SBP), DBP, DM, Smoking, ECG-LVH if available.

The European Score
The European Heart SCORE model constitutes the basis for national guidelines for primary prevention and treatment of cardiovascular disease (CVD) in several European countries. The model estimates individuals' 10-year CVD mortality risks from age, sex, smoking status, systolic blood pressure, and total cholesterol level. The SCORE model, however, is not mathematically consistent and does not estimate all-cause mortality [19].

Finrisk Model
It is based on its European origin. It included Diabetes and LDL [20]. The FINRISK model was derived from the occurrence of acute myocardial infarction or coronary heart disease death during a 10 year follow-up of populations in eastern and southwestern Finland, based on the 1982 and 1987 cohorts.

Berto Model
The model from Berto and co-workers starts with healthy individuals and model their life until different coronary heartdiseases (CHD) or death by other causes [21]. This model does not go any further after the first CHD event.

CDC Model
CDC (Centers for disease control and prevention) diabetes cost-effectiveness group made a Markov model which models whether an individual gets CHD or continues to stay "normal". In contrary to e.g. Berto's model, this model doesn't stop at first event, but also models that one can stay in the health states "angina" and "history of CA/MI" for several years.

Huse Model
Huse and co-workers have, as CDC, modeled up to all individuals are dead [22]. This is however a model which evaluates the difference between different statins. In this model it is possible to start and stop giving medications during the process, which is rather uncommon for models like these.

Nyman et al. Model
It models the whole process up to death [23]. In this model, all individuals who survive the first year after a cardiovascular event will end up in a "chronic state" and be there until their death.

Weinstein's Model
This model is possibly the most extensive of those mentioned here [24]. This model has five health states; persons without CHD, persons with new CHD, persons with CHD, persons who survived up to the age of 85 years, and death. Hence this model has two different absorbing states; survive 85 and death. Individuals in the state "new CHD" will only be there the first year, and after that transfer to "CHD", "survive until 85" or die.

CANFIS Model
An approach based on coactive neuro-fuzzy inference system (CANFIS) was presented for prediction of heart disease. The proposed CANFIS model combined the neural network adaptive capabilities and the fuzzy logic qualitative approach which is then integrated with genetic algorithm to diagnose the presence of the disease. Coactive Neuro-fuzzy modeling was proposed as a dependable and robust method developed to identify a nonlinear relationship and mapping between the different attributes. It has been shown that of Genetic Algorithm (GA) is a very useful technique for auto-tuning of the CANFIS parameters and selection of optimal feature set [25].

Data Description -Patients Data
The data for this study has been collected from various Hospitals of 689 patients, who have symptoms of heart disease. All these cases are analyzed after careful scrutiny with the help of the Physicians. Table 1 shows the various input parameters for the prediction of heart disease. For training and evaluation purpose we have carefully predicted the level of heart disease by taking the help of Cardiologist/ Physician. The data consists of patients' record with doctor's predictions/ diagnosis. The whole dataset is divided into training, validation and testing as shown in Table 2.

Proposed Cardiovascular Disease Prediction Systems
This System provides the comprehensive predictive tool for predicting cardiovascular disease; this system predicts the cardiovascular disease in early stage by clinicians, doctors, pathologists as well as by general public also, may predict the chances of having the cardiovascular diseases using the important Clinical tests caused for cardiovascular disease, so that the clinicians may fulfill their promises and

Quantum Neural Network Used for Cardiovascular Disease Prediction System
Quantum Neural Network based on multi-level

Algorithm for Training the Quantum Neural Network
The QNN used for our proposed system consists of input, output and hidden units, as shown in Fig. 2. Here, only one hidden layer has been used. In our proposed system, the QNN's every node of hidden layer represents only three sub states in itself with the difference of quantum interval θ r with quantum level r. Here more than three sub states may be possible in every node of hidden layer, but with every additional sub state the computational complexity increases, but for our proposed system, with three sub states the performance of QNN is optimum with significant accuracy. Let us assume that n s denotes the number of grades or excitation levels, η is learning rate which is a small random value, δ k is error rate of output layer and δ j error rate of hidden layer. Where n i denotes the input to the input layer. O j and O k denotes the output of hidden and output layer, respectively. The weights between input and hidden layers are denoted by W ij and the weights between hidden and output layers are denoted by W kj . The initial weights are small random numbers and t denotes Target value. First apply the inputs to the network and calculate the output [29,30,31].
Given R training pairs {n 1 , t 1 ; n 2 , t 2 ; n R , t R } Where n i (J X 1) is input and t i (K X 1) is target values for given inputs, for the present quantum neural network as shown in Fig. 1.
The error signal term of the output (δ k ) and hidden layers (δ j ) are written as, The simple sigmoid function has been used as the activation function from input to hidden layer and is expressed as: The sigmoid function with various graded levels has been used as the activation function for each hidden neuron i.e. from hidden to output layer is expressed as follows: Here the initial value of θ r is initialized by zero, then apply Eq1 to 6 and iterate this process several times upto i iterations until it reaches the lowest possible error and then apply Eq 7.
After i iterations when minimum possible error is obtained, then increase in quantum interval by very small quantum interval ∆θ, for this proposed system we have used the value of ∆θ is 0.25.
Repeat this step until δ k is decreased to an acceptable accuracy.

RESULTS AND DISCUSSION
The proposed system predicts the rate of heart disease or the percentage rate of disease on the basis of Clinical risk factors; it also predicts the future condition in advance. The Proposed system is based on Quantum Neural Network, means it is predictive as well as adopts the predictive knowledge based on the past experience i.e. on the basis of past data it learns, and gradually maximizes the accuracy level of prediction. We have categorized the risk level into four groups according to percentage level as Normal, Low Risk Range, Intermediate risk range and highest risk range. The percentage for each level is shown in Table 3.
The results of our experimental analysis for significant patterns for cardiovascular disease prediction using Clinical risk factors are presented in this section. In this work, we have only used Clinical risk factors significant to cardiovascular Disease. After analysis of patients' data, the patients may be categorized on the basis of reference range. As shown in the Fig. 1, the total numbers of patients are distributed percentage wise patient distribution lies in Lower, Average and High Level Reference rangeof Diabetes Mellitus (DM), Low Density Lipoprotein (LDL), Triglycerides (TG) and High Density Lipoprotein (HDL).With the help of the dataset, the patterns significant to the cardiovascular disease prediction are extracted. The dataset is divided into training set, validation set and test set as shown in Table 2.  The Quantum Neural Network is trained with the selected significant patterns. The architecture of the quantum neural used in this study is having 4 input nodes, 15 hidden nodes and 1 output nodes as shown in Fig. 2.
The Quantum Neural Network is trained with the above mentioned number of nodes and accuracy graph is shown in the Fig. 3 and its performance graph is shown in Fig. 4. The number of input nodes is determined by the clinical parameters, the number of hidden nodes is determined through tuning; and the number of output node is The Quantum Neural Network is trained with the selected significant patterns. The architecture of the quantum neural network used in this study is having 4 input nodes, 15 hidden nodes and 1 output nodes as shown in The Quantum Neural Network is trained with the above mentioned number of nodes and accuracy graph is shown in the Fig. 3 and its performance graph is shown in Fig. 4. The number of input nodes is determined by the clinical parameters, nodes is determined through tuning; and the number of output node is represented as percentage level of cardiovascular disease.
The proposed work shows that the prediction of risk from cardiovascular diseases gives best results on the dataset used. The re by this system has been evaluated and validated on data of patients with the Doctor's diagnosis (predictions). When the QNN is trained and tested after optimizing the input parameters, the overall predictive accuracy obtained is 96.97%. Data Showing Prediction accuracy of Random Testing on different Experimental values been given in Table 4.

CONCLUSION
The proposed work shows that the prediction of risk from cardiovascular diseases gives best results based on clinical factors. The result generated by this system has been evaluated and validated on data of patients with the Doctor's diagnosis (predictions). This proposed system will help the doctors to plan for a better medication and provide the patient with early diagnosis as it performs reasonably well even without retraining. The QNN has been trained and tested after optimizing the input parameters. The optimization has been done by selecting minimum distorted data set of input parameters. The overall predictive accuracy obtained was 96.97%.Data Showing Prediction accuracy of Random Testing on different Experimental values has been given in (Table 4). Hence the proposed system will work as a significant tool for doctors/ practitioners.