Blood vessel disease and heart disease also called cardiovascular disease (CVD) or heart illness. Cardiac disease is the world's leading cause of morbidity and mortality. It includes stroke, heart arrhythmia, heart failure, carditis, congenital heart disease, hypertensive heart disease, valvular heart disease, venous thrombosis, peripheral artery disease. It is estimated that, if predicted in early phases, 90 percent of CVDs are preventable. It is therefore essential to identify such computational illnesses (Anika, Kaur, 2017). Therefore, it is not always simple to identify heart disease because for an early forecast it needs expert understanding or feelings about symptoms of heart failure. Thus, by precise diagnosis through biochemical exams and suitable therapy, the number of fatalities of patients with cardiovascular disease can be decreased. In the healthcare sector, there is an enormous quantity of data available. All this information is stored in large electronic medical recording databases. Drowning in information but hunger for understanding and providing a precise diagnosis of cardiovascular disease has become a huge challenge for clinics. Data may be studied for assessment purposes, but hospitals have not yet adequately explored them. It was possible to develop the mining of medical records for analytical purposes to guide and assist the clinical decision-making process (Miranda et al., 2016).
The medical imaging system has developed significantly over the last twenty years. Technology advances such as digital photon emission tomography (PET), ultrasound (US), multi-slice computed tomography (CT), parallel magnetic resonance imaging (MRI), these technologies enabled better images with higher resolution. A wealth of information is contained by today's medical images, as important information is hidden in the voxels or pixels. Therefore, software, workstations, and solutions are provided by producing medical imaging system for visualizing, archiving and analyzing pictures within the context of sickness areas like medical specialty, cardiology, neurology with the aim to support designation, screening, treatment and follow-up examinations. Medical Image Analysis provides a forum for the dissemination of new research results in the field of medical and biological-image analysis, with special emphasis on efforts related to the applications of computer vision, virtual reality, and robotics to biomedical imaging problems (Weese, Lorenz, 2016). The significant objective of medical image analysis is to convert the raw pictures into a quantifiable form for reasoning, assessment, and indexing. This task is challenging for medical imaging research, medical practice and the healthcare sector because the quantification of medical image content is complex and not a problem solved. More importantly, in real-time few techniques are used to analyze the big picture database, let alone integrate user requirements. Because the development of large-scale medical picture analysis algorithms has largely lagged the growing quality of medical images and the imaging itself, there is an urgent need to create innovative and integrated frameworks to enable robust and timely medical imaging analysis, disease characterization and search in appropriate databases (Zhang, Metaxas, 2016).
Millions of individuals get cardiac disease every year and cardiac disease is the largest murderer of both males and females in the United States and around the globe. The World Health Organization (WHO) analyzed that heart disease causes twelve million fatalities globally. In nearly every 34 seconds, one individual in the globe is killed by heart disease.
Medical diagnosis plays a crucial and yet complex role, which requires to be effectively and correctly performed. Suitable computer-based data and decision support should be helped to decrease the cost of clinical testing. Data-driven machine learning (ML) methods could enhance risk prediction efficiency by agnostically finding new risk predictors and studying the complicated relationships between them (Patel, Upadhyay, Patel, 2016). Machine learning (ML) is an artificial intelligence domain which includes building algorithms that can learn from experience. The way ML algorithms function is by detecting hidden patterns and building models in the input dataset. They can then create precise predictions for fresh algorithm datasets that are completely new. Through learning, the machine thus became smarter; so, it can recognize patterns that are very difficult or impossible for people to detect on their own. Large datasets can be operated by ML algorithms to make predictions and decisions (Janabi, Qutqut, Hijjawi, 2018).
2.1 In medical imaging, there are various AI methods that are robust to extract unseen, predictive and actionable data from comprehensive databases. This information can be used to obtain helpful insights using multiple machine learning techniques and data mining is an interesting field of machine learning and therefore very well able to solve this sort of issue (Ramalingam, Dandapath, Raja, 2018). AI technology has been already used in the medical field for AI imaging diagnosis support systems in radiography, CT scanning (computed tomography), and MRI (magnetic resonance imaging), all of which use images to detect pathological changes. To employ machine learning, a much-talked-about application of deep learning, one must collect an enough (over 100,000 pieces) of both normal and abnormal data for learning to enable a high recognition accuracy. Let first explain these technical terms.
CT scanning: CT scanning is a test used for viewing the heart and coronary arteries. It utilizes a mixture of X-ray and computer technology to identify blockages in the artery walls and/or calcification-the factors contributing to angina.
MRI: MRI is a sort of non-invasive test that utilizes radiofrequency waves and magnetic fields to get a comprehensive image of the human body's organs and structures. It examines both the heart disease and the regions impacted by the stroke in a perfect way. In addition, MRI can assist physicians diagnose many associated heart conditions.
ULTRASOUND: Echocardiogram, cardiac echo, or transthoracic echo (TTE) is also referred to as cardiac ultrasound. To generate a moving image of it, it utilizes ultrasonic waves that bounce off the core. It allows doctors to see the heart in movement, including the activity of beating. An echocardiogram offers more data and visualization than a picture with x-rays.
2.2 MACHINE LEARNING TECHNIQUES
Machine learning (ML) is an artificial intelligence domain which includes building algorithms that can learn from experience. The way ML algorithms function is by detecting hidden patterns and building models in the input dataset. They can then create precise predictions for fresh algorithm datasets that are completely new. through learning, the machine became smarter, so it can recognize patterns that are very difficult or impossible for people to detect on their own. ML algorithms and methods can work with big datasets and make predictions and decisions.
3. REVIEWED APPROACHES BASED ON AI TECHNIQUE FOR HEART DISEASE DETECTION (Machine learning)
NaÃ¯ve bayes: Medhekar et al. in 2013 proposed a scheme for categorizing information into five classifications using the classifier Naive Bayes. No, low, average, high and very high categories. The scheme predicts input information on the likelihood of heart disease. The UCI heart disease dataset shown in table 1 is the dataset used for practice and testing. The scheme was 88.96 percent accurate.
Vembandasamy et al. in 2015 used Naive Bayes classifier to diagnose heart disease existence or absence. The dataset used in the study is acquired from one of Chennai's leading diabetic research institutes with records of approximately 500 patients and 11 characteristics (including diagnosis). To apply the Naive Bayes classifier, the Waikato Environment for Knowledge Analysis (WEKA) instrument, a collection of ML algorithms, is used. Their study work was accurate at 86,4198 percent.
J, U, D S. in 2011 they used a naÃ¯ve Bayes strategy for prediction of heart illness with a precision of 86.53% using 22 predictor characteristics. Two biochemical characteristics were used in this study, namely cholesterol and blood sugar. Two techniques have been used to evaluate this study. First, precision, sensitivity, and specificity were calculated and for all risk concentrations, each value was above 80 percent. Second, the model was evaluated by a cardiologist and an internist through an assessment session.
Artificial Neural Network: Elalfi, Eisa, Ahmed in 2013 presents an image processing-based artificial neural network for the diagnosis of heart valve diseases they used a neural network to conduct image classification; it has three layers; input, hidden, and output layer. The technique suggested was applied using MATLAB. Mansoura University Hospitals provide the picture database in the experiment. In order to evaluate the performance of this work, precision, recall, and Accuracy were used. Altogether, the total accuracy rate was 93.75%.
Arabasadi et al. in 2017 propose a highly hybrid method for the diagnoses of coronary artery diseases. The proposed method can increase the performance of the neural network by approximately 10% through enhancing its initial weights using a genetic algorithm. The research used Z-Alizadeh Sani dataset containing information on 303 patients, 216 of whom suffered from CAD. They achieve accuracy, specificity, sensitivity rates of 93.85%,92%, and 97% respectively.
Danger and Apte in 2012 used ANN develop a prediction scheme for heart disease (HDPS) to predict the existence or lack of heart disease in patients. The Cleveland heart disease data set was used to train the algorithm and the Starlog data set was used to test; Both have been acquired from the UCI repository and have 13 medical characteristics. The experimental instrument used is the WEKA instrument. The findings showed that using the thirteen characteristics offered 99.25 percent precision while using the fifteen characteristics offered almost 100 percent precision to predict the disease.
Awan, Riaz, Khan, in 2018 used ANN with the KDD model to diagnose heart disease. They have taken data from heart disease diagnosis medical reports used “cleaver land heart data” taken from UCI repository. For research 13 attributes were selected. Weka is used as a tool for the implementation of the methodology. The accuracy is calculated and visualized such as ANN gives 94.7% but with Principle Component Analysis (PCA) accuracy rate improves to 97.7%.
Kirmani, Ansarullah. 2016, This study work explores the outcomes after applying a variety of methods to various kinds of Decision Trees to achieve better efficiency in the diagnosis of heart disease. The awareness, specificity, and precision are calculated to assess the efficiency of the alternative decision trees. This study work suggests a model that works better in the diagnosis of heart disease than the J48 Decision Tree and Bagging algorithm. The information used in this research is derived from the information set available for the Cleveland Clinic Foundation Heart disease. There are 76 raw characteristics in the information set. All the tests released, however, refer only to 13 characteristics. The highest accuracy achieved is 85.3% by the Equal Frequency discretization Gini Index Decision Tree.
Maji S., Arora S. (2019), In this paper, the hybridization technique is proposed in which decision tree and artificial neural network classifiers are hybridized for better performance of prediction of heart disease. This is done using WEKA. To validate the performance of the proposed algorithm, a tenfold validation test is performed on the dataset of heart disease patients which is taken from UCI repository. The accuracy, sensitivity, and specificity of the individual classifier and hybrid technique are analysed.
Pandey et al. 2013, In this paper cardiovascular forecast model, is created that can help medical experts in predicting the status of cardiovascular illness based on patient clinical information. They select 14 attributes having clinical features. Then develop a predictive model based on J48 decision tree to classify heart illness against unpruned, pruned with decreased error pruning strategy. Finally, Pruned J48 Decision Tree's precision with Reduced Error Pruning Approach is better than Pruned and Unpruned easy method. The outcome has been that showing that fasting blood sugar is the most significant characteristic that provides better classification against the other characteristics but does not give better precision.
Support Vector Machine: Ghumbre, Patil, Ghatolin in 2011, this article investigates the implementation of artificial intelligence in the diagnosis of typical heart disease. A smart system-based support vector machine is provided for diagnosis in this study along with a radial base function network. An expert system based on clinical symptoms is used to determine which sort of heart disease can occur for a patient, whether it is a heart attack. The support vector machine with a minimal sequential optimization algorithm is implemented to the information set of Indian patients. Then the network design of the Radial Basis Function (RBF) trained by the Orthogonal Least Square algorithm (OLS) is implemented for predictions to the same information set. Results acquired show that to diagnose heart disease, a support vector machine can be used effectively. The results suggest the function of efficient diagnosis and the benefits of information coaching on the automatic medical diagnosis scheme based on machine learning.
Wiharto et al. in 2015, The precision of the UCI dataset SVM algorithm kinds for the diagnosis of heart disease were studied. The research included several kinds of SVM, including Binary Tree Support Vector Machine (BTSVM), One-Against-One (OAO), One-Against-All (OAA), Decision Direct Acyclic Graph (DDAG) and Exhaustive Output Error Correction Code (ECOC). The accuracy of the types of UCI data set SVM algorithm for heart disease diagnosis has been researched. Using a min-max scaler, the data set was first pre-processed. The next phase was to train the data set algorithm using the above-described SVM algorithms. BTSVM conducted better than the other algorithms in the performance evaluation with a general precision of 61.86 percent.
K-nearest Neighbour: Shouman, Turner, Stocker in 2011, This article explores the application of KNN in the diagnosis of heart disease to assist healthcare practitioners. It also explores whether the integration of voting with KNN can improve its precision in the diagnosis of patients with heart disease. The information used in this research is the heart disease benchmark of the Cleveland Clinic Foundation. The findings indicate that applying KNN in the diagnosis of patients with heart disease could attain greater precision than neural network ensemble. The findings also indicate that applying voting in the diagnosis of heart disease could not improve the precision of KNN.
Jabbar, in 2017, This paper addressed the prediction of heart disease based on PSO and KNN. Our approach uses KNN as a classifier to reduce the misclassification rate. This paper also investigates PSO based feature selection measures to select a small number of features and to improve the classification performance. To predict heart disease the dataset containing 270 instances collected from UCI repository. WEKA is used as the main package. Experimental results show that the algorithm performs very well with 100% accuracy with PSO as feature selection.
E, K, P, Kumar, K in 2016, In this article we provided an approach to cardiac disease prediction using K nearest Neighbour and K Means execution algorithms. The suggested scheme group health-related characteristics into a number of clusters and calculate the cluster centroid value using KNN algorithm, as well as calculate the cluster centroid value using K implies algorithm. These values help forecast a person's heart disease.
Hybrid approach: Khateeb and Usman in 2017, experimented with different classification algorithms such as decision tree Naive Bayes, KNN and UCI Cleveland data set bagging techniques. The research has been split into six instances, and each classifier calculates the precision for each situation. In case 1, without feature reduction, all the classifiers were applied to the dataset. In case 2, function decrease was used where only chosen seven attributes, the most important for heart disease diagnosis, was chosen instead of using all 14 characteristics in the dataset. n case 3, only the most general characteristics such as era, gender and blood sugar resting have been removed. In case 4, WEKA instrument resampled the dataset and only used the seven most important characteristics. The resampling improved each classifier's precision. In case 5, all 14 characteristics were resampled. The Synthetic Minority Over-Sampling Technique (SMOTE) was finally implemented in WEKA instrument in case 6. The best outcome was to use KNN on case 5, which produced a precision of 79.20 percent.
Venkatalakshmi and Shivsankar in 2014 Included a comparison between Naive Bayes and Decision Tree to determine which heart disease forecast has the greatest precision. The dataset used was the dataset for heart disease UCI. The experiment was conducted using the WEKA tool and showed a precision for Naive Bayes and Decision Tree of 85.03 percent and 84.01 percent respectively. The document suggested using a MATLAB genetic algorithm to decrease the number of characteristics for the future job before feeding the dataset into the WEKA instrument.
5.CHALLENGES AND FUTURE RESEARCH
The present paper shows how machine learning is used in medical image analysis to predict heart diseases on time, however, it may not be easily applicable to diseases with variable abnormal areas or other imaging modalities. Also, it is important to consider where this approach could be best applied: would it be for screening the general population in the primary healthcare setting or in aiding ophthalmologists in making diagnoses in tertiary care settings? and more generally, future studies might address challenges in medical imaging, such as how and/or when machines and human adjudication differ, and design methods that quantitatively and qualitatively assess and explain sources of error for both humans and machines medical imaging analysis, however, still have questions that remain unanswered. Thus, it is critical for machine learning and medical communities to collaborate closely not only to facilitate the development and validation of machine learning, but also to strategically deploy these technologies for patient care.
According to Alexander et al., in 2018, machine learning methods seem to take over the field and in image-based diagnosis, disease prognosis, and risk assessment are progressively effective. In order to unlock their complete potential, many science and practical issues still need to be resolved, including how to train powerful models on the small data.t. How to enhance information access, how to best use the picture structure and characteristics of medical imaging information when developing our models, how to interpret outcomes and how to apply them in clinical practice.
Machine learning methods seem to take over the field and are progressively effective in imaging, illness prognosis and risk evaluation. Many science and practical issues still need to be resolved in order to unlock their complete potential, including how to train powerful models on small information, how to enhance data access, how to make best use of the picture structure and specific properties of medical imaging information when designing our models, how to interpret outcomes, and how to apply them in clinical practice.