IoHT with Wearable Devices–Based Feature Extraction and a Deep Neural Networks Classification Model for Heart Disease Diagnosis
Introduction
Internet of Things (IoT) concepts are employed in diverse fields transforming the way that business processes are made [1]. Health informatics is a promising multidisciplinary domain that focuses on employing information engineering concepts to healthcare. The information usually originates from a diversity of sources such as healthcare information technology systems, but lately it is being saved in distinct IoT devices [2, 3]. The application of IoT concepts is becoming the norm, giving rise to the Internet of Health Things (IoHT). In general, heart disease (HD) is said to be a more serious disease that affects the function of the human heart and tends to increase the chance of a coronary artery or lower blood vessel event. Such complications lead to a heart attack or stroke. Based on the study of [1], around 610,000 people are affected by HD in the United States. Although HD affects males and females, males are more positive for heart attacks. The study reveals that the signs of HD [2] are chest tightness, pain, pressure, breathing issues, leg chills, neck pain, abdominal pain, tachycardia, light headedness, bradycardia, dizziness, syncope, change in skin color, leg swelling, weight loss, and fatigue. Sometimes, the symptoms differ based on the nature of HD such as arrhythmia, myocardia, heart attack, congenital HD. mitral regurgitation, and dilated cardiomyopathy. Some of the risk factors involved in HD are age, genetics, smoking, sex habits, drug abuse, higher cholesterol, high BP, external inactivity, obesity, diabetes, stress, and poor diet and hygiene. The severity of HD requires the disease analyzing process to be focused on diagnosing at an early stage.
While undergoing the screening process, physicians take into account the level of blood glucose, cholesterol test, BP test, electrocardiography (ECG), ultrasound, cardiac computer tomography (CT) calcium rate, and stress test. Therefore, the screening task [4] requires a massive time interval for manual intervention. There are various automated realizing models employed to find the function of human heart and pattern modification. Additionally, a few data mining (DM) methods, machine learning (ML), and artificial intelligence (AI) methodologies [5] were utilized to perform the heart information. For enhancing the model’s function, an automated system extracts gathered data and removes the noise. Furthermore, diverse filtering approaches [6] such as normalization mean filters have been applied for processing the data [7]. Therefore, these automated models require a massive volume of data, which results in system complications. This difficulty tends to minimize efficiency in detecting HD. Hence, the automatic system is combined with a smarter device according to IoT [8] for collecting patient details.
Here, IoT is comprised of a set of devices and sensors that are employed to collect data from a specific platform. The efficiency of a sensor device was developed by Kevin Ashton at MIT campus. The tool applies the radio frequency ID and P and G sensor managing device [9] for gathering data from the human body. The Io- centric communication task [10] enhances the total experience of a patient and the common efficiency of this task. Here, the advantages of IoT devices are harnessed by combining with an automated disease detecting model for analyzing HD. At the time of processing this operation, the system might predict the HD features with minimum accuracy because it has poor training, learning, and examining processes. Hence, major contributions of this method are applied for improvising HD prediction value under the application of a massive amount of data, lower time consumption at the time of detecting HD, and assuring reduced false classification value during the detection of HD.
Broad research has been carried out for incrementing the diagnosis of HD. In [11], cardiovascular HD has been forecasted with the help of an optimized genetic algorithm (GA) with fuzzy recurrent network. Also, it is employed with a benchmark database to estimate the HD. The patient data is computed by applying data processing methodologies along with fuzzy-rule based technique. The recurrent network segments that provide input detects the attained results for effective training, which can be accomplished by using GA. Even though the developed model attains maximum recognition value, which tends to the production design of fuzzy classification rules, it does not provide exact accuracy for a massive amount of heart data. In [12], optimal neural networks (NN) were employed for finding HD-based information. It helps collect different clinical as well as heart data, which is computed successfully to remove the noise. Consequently, to diagnose the variations in data, it is provided into feed forward NN (FFNN). While processing the analyzing task, the parameters undergo optimization under the application of genetic operators, which reduce analyzing difficulty. Therefore, it is effectively deploying a method that predicts HD with maximum accuracy than conventional classification models like support vector machine (SVM) and к-nearest neighbor (K-NN).
In [13], the swarm optimized convolution neural network (CNN), with SVM, is applied in recognizing HD. While implementing, the deviation in a kidney, a test has been carried out under the application of chronic disease-based data such as saliva, ammonia, and concentration of urea. The data is computed using SVM along with a swarm intelligence (SI) training model. Furthermore, the affected features undergo classification by applying CNN, which analyzes HD with maximum accuracy.
Therefore, failure explains the efficiency of the feature selection (FS) process that produces the difficulty and improves the processing duration.
In [14], the efficiency of different intelligent health care modules, such as data transmission framework IoT, big data, and smart decision-making processes that are applied in disease diagnosis, is discussed. The gathered data undergo investigation by employing various ML techniques, stochastic models, and evolutionary approaches. The combination of these models leads to automated disease analysis for improving the total diagnostic process. Besides, even though it is applied in diagnosing diseases like diabetes and HD, it has failed to mention the way of capturing data from a subject, and tends to maintain the data analyzing process.
In [15], an IoT-centric HD recognition module is deployed based on the ML technique. The collected data has been computed with the application of SVM, which accurately classifies normal as well as a cardiovascular disease with higher accuracy. Therefore, the model collects details about HD such as body temperature, BP, heartbeat, and humidity level using IoT devices. It is repeated for obtaining accurate values regarding HD. Also, it is capable of recognizing HD with reduced time but fails to control the real-time heart data in the case of a higher amount of data in a technique. In [16], HD has been detected by employing the Cleveland clinic database with the help of particle-optimized feed forward back propagated NN (BPNN). The gathered features are examined based on selecting optimal features to particle position as well as velocity. Such features can be determined under the application of feed forward BPNN to attain the accurate features for abnormal HD. This method analyzes the HD with higher accuracy and reduced cost.
In [17], mortal HD predicting is deployed by using a particle swarm-optimized radial basis function network. The authors also look at HD details that analyze the heart features concerning nonlinearity as well as linearity. Also, extracted features have been computed, and selected features from abnormal HD features undergo classification. Hence, it is based on heart rate inter-beat (RR) interval to examine heart value with partial accuracy. Therefore, diverse automated models are capable of analyzing HD, but it requires a larger amount of heart details to improve accuracy. The newly presented method is applied for analyzing deviations in heart features with lower complexity as well as higher accuracy.
This chapter presents a new IoT wearable-based heart disease diagnosis model. The proposed model gathers the patient details and transmits the data to the health care center. Then, the feature process takes place followed by a deep neural network (DNN)-based classification. The proposed model will effectively predict the presence of heart disease from the data gathered by the IoT wearable. The effectiveness of the proposed model has been tested using a benchmark data set. The results indicated that the proposed model outperforms the existing models in a significant way.
