International Journal of Advance Engineering and Research Development SURVEY IN ARTIFICIAL NEURAL NETWORKS IN MYOCARDIAL INFARCTION

Transcription

1 Scientific Journal of Impact Factor (SJIF): 5.71 International Journal of Advance Engineering and Research Development Volume 5, Issue 03, March e-issn (O): p-issn (P): SURVEY IN ARTIFICIAL NEURAL NETWORKS IN MYOCARDIAL INFARCTION Dr. Veena. S 1, Nerisai. M. V 2, Vignesh. P 3 1 Professor, Department of Computer Science and Engineering 2 Student, Department of Computer Science and Engineering 3 Student, Department of Computer Science and Engineering S. A. Engineeering College Abstract- The most important branches of computer science are the Artificial Intelligence, which increases the capability of analyzing complex medical data. It contains more potential which has to be exploited by the human beings for sake of them. It helps to diagnose, to treat and to predict the outcome in many clinical scenarios as it have good relationship with a data set. Since accuracy and efficiency are the two principles needed for the better treatment to save the life of a human community, the proficiency in Artificial Intelligence is an essential high time need in all the fields of medicine. Based on this, the present paper tries to explore the various outlooks of AI in medicinal field. Keywords- Artificial Intelligence, Medicine, Neural Networks, neural topology I. INTRODUCTION Aristotle s attempt to formalize right thinking (logic) through a three part deductive reasoning, inspires many modern era workers to study on the operation of mind helped to establish contemporary logical thinking. Computer programs, makes it function in such a way that make the people to think as intelligent is called Artificial Intelligence system. The technology (Artificial Intelligence) that has been created to assist human beings to perform various activities with more accuracy and efficiency. The development of AI is associated with human intelligence like learning, reasoning and problem solving. Though the process for developing AI began in the first half of the 20th century, the creation of The Logic theorist designed by Newell and Simon in the year 1955 was the first ahead step towards the modern AI. Earlier it was related to engineering. Later the applications are extended towards all departments of science and technology including medicine field. The application of AI technology in medicine was first successfully investigated by Gunn in 1976 and explored the possibility of diagnosing acute abdominal pain with computer analysis. In the last three decades we have seen a surge in the interest in medical AI. The ultimate problem of resource scarcity in India is the health care issues. The country requires more medical facilities and expertise but it needs more time and money to develop. As they are not easily obtainable, we should consider different ways to increase access to the existing resources in an inexpensive and effective way. A company named Tricog Health is a startup company which was handpicked by GE s health care accelerator program for cloud-based cardiac diagnosis problem. Nearly 26% of adult deaths in 2003 are escalated to 32% in 2013 due to Coronary heart disease as they are increasingly prevalent in India. This company uses specialized Artificial Intelligence to process the collected physiological data and ECGs from medical devices in real time and give cardiologist a diagnosis. After that, the cardiologist reviews the given diagnosis and recommends next steps to the nurse using the mobile app created by Tricog. A few specialists can diagnose over 20,000 patients using Tricog s AI engine. Now-a-days we are facing many challenges in acquiring, analyzing and applying the large amount of data and knowledge to solve complex clinical problems. Modern AI extends its helping hands for the formulation of a diagnosis, making therapeutic decisions and the prediction of outcome. They also face challenges like poor safety record, environmental pollution, unreliability capacity problems and wasted energy. The challenge is the fact that transportation systems are All rights Reserved 1568

2 complex systems which involves large number of components and different parties, each of them having different and conflicting objectives often. Artificial Neural Networks (ANN), evolutionary computation and hybrid intelligent systems are very much assisting and supporting the health care workers on their daily activities. India has the capacity to provide the huge amounts of data required to improve the accuracy of these algorithms and empower both startups and large companies to help solve health care problems around the world with 1.3 billion citizens. So, by utilizing this potential works to be triggered on this field which forms the basis for the present work. II. ARTIFICIAL NEURAL NETWORKS (ANN):- Dr. Robert Hecht- Nelson who is the inventor of neurocomputer, defines a neural network as a computing system made up of a number of a simple, highly linked processing elements, which process information by their dynamic state response to external in puts. In the past three decades, ANN becomes more popular in the field of medicine. ANN are nothing but computational analytical tool whose inspiration came from biological nervous system. Just like interconnected neurons in the brain to receive, store, recall, analyze the data and give information for performance, ANN has capable of performing parallel computations for data processing. It learns from previous experiences stored as data and perform accordingly in response patterns accurately, so the researchers are interested to solve many clinical problems by applying them in it. It helps for diagnosis, treatment and predicting outcome in many critical clinical situations. In 1990, one of the first researchers to explore the clinical potentials of ANNs newly developed neuronal network model by Baxt, accurately diagnosed acute myocardial infarction and latter possibly validated his work with similar accuracy (Ledley and Lusted, 1959). From this onwards, the application of ANN in medicine field becomes most wanted one. ANN has been used in the clinical diagnosis, image analysis in radiology and histopathology, data interpretation in intensive care setting and waveform analysis. Prost A sure Index, neuronal network derived classification algorithm was developed by Stamey etal (1996). With 90% diagnosed accuracy validated for many years. Then, it was extended to surgically relevant applications including abdomen pain, appendicitis, common bile stone problems, glaucoma, back pain and to classify prostates as benign or malignant. PAPNET, another ANN, computerized automated screening system based neuronal networks assisted cytological workers as it helps to screen, breast cancer, gastric cancer, thyroid cancer, oral cancer, urothelial cells, pleural and peritoneal effusion cytology. Pattern identification ability of ANN makes to analyse ECG pattern to diagnose myocardial infarction, atrial fibrillation and neutricular arrhythmias. Analysis of EEG (electroencephalogram) also possible by it. III. SOME TYPES OF ARTIFICIAL NEURAL NETWORKS:- a)multi-layer perceptron: In this study, the Artificial Neural Network classifier used for the multi-layer perceptron (MLP) network is a standard feedforward system which contains a single hidden layer and a back propagation training algorithm. Every single input neuron is attached to a hidden neuron, and all of them are subsequently connected to the output neuron. A numerical input is received for each input neuron from all input features, which are normalized within an interval of 0-1. The values are then multiplied by the connection weights, that represent the relative influences between the neurons from the layer one to the next layer. All the multiplications are added and sent through the network. The function that is most frequently activated is the sigmoid function. It simulates the all-or-none behavior of biological neurons. The output of hidden neurons is multiplied with appropriate connection weights and fed to the last decision-maker neuron. The output neuron also performs identical calculations which produce the output of the network that s final. The optimal number of hidden neurons that was selected would result in a predictive network with maximal sensitivity and All rights Reserved 1569

3 b) Network model of Radial basis function: The RBF network also has a feed-forward structure like multi-layer perceptron consists of a single hidden layer containing J locally tuned units are fully interconnected to an output layer which contains a decision-maker neuron. It should be observed that the hidden neurons are more biased to the central data located around the centers. All hidden units receive the n- dimensional real value input vector X in parallel. The RBF networks can also be used for reversion and pattern classification tasks. Its network has 3 layers. Layer one consists of receptor neurons and the hidden layer usually contains several neurons with a Gaussian activation function. c) Neural Network topologies: This section mostly focuses on the pattern of connections between the propagation and the units of data. For this pattern of connections, the main distinction we can make is between: In Feed-forward neural networks, the data flow from input to output units is strictly feed-forward. Over multiple units can be extended in data processing. But no feedback connections are present, that is, the connections extending from outputs of units to inputs of units in the same layer or previous layers. d)recurrent neural network: It contains feedback connections. In opposite to feed-forward networks, the dynamical properties of the network are important. The activation values of the units undergo a relaxation process in some cases such that the neural network will evolve to a stable state in which these activations do not change anymore. The change of the activation values of the output neurons are significant in other applications. Some examples for feed-forward neural networks are the Perceptron and Adaline. Anderson presented the examples of recurrent networks such as (Anderson, 1977), Kohonen (Kohonen, 1977), and Hopfield (Hopfield, 1982). Area: 2000cm 2 Figure 1. Types of Neural Networks Weight: 1,5 kg Covering the hemispheres of the cerebral cortex contains neurons: The number of connections between cells: The cells send and receive signals, the speed of operation= operations / sec The neural network is a simplified model of the brain IV. NEURAL NETWORKS IN MEDICINE FIELD:- At the moment, the research on ANN in medicine field is on modelling parts of the human body and in recognising diseases from various scans like cardiograms, CAT scans, ultrasonic scans, etc. Neural networks can recognize diseases using scans as there is no need for providing a specific algorithm on how to identify the disease. Neural networks are learned by example so the information of how to recognise the disease is All rights Reserved 1570

4 needed. What is required is a set of examples that represents all the variations of the disease. All these examples are should to be selected very carefully if the system should perform reliably. V. FUNDAMENTAL TRAIT OF INTELLIGENCE IS THE ABILITY TO LEARN:- Even though a precise definition of learning is difficult to formulate, learning process in ANN context can be viewed as the problem of updating network architecture and connection weights so that a network can perform a specific task efficiently. Usually the network must be learned by the connection weights from available training patterns. The performance is improved by updating the weights in the network. The ability to automatically learn from examples makes ANN more attractive and exciting. ANN learn underlying rules (like input-output relationships) from the given collection of representative examples. It is one of the major advantages of neural networks over traditional expert systems. VI. REASONS FOR MISCLASSIFICATION:- Some ECGS were misclassified by the ANN and correctly classified by the electrocardiographer. To train the neural networks (in this study), a relatively small number of input variables was used. A network that is fed with many input variables requires many examples in the training set. The number of training examples required for appropriate training is 10 times the total no of inter neuron connections in the neural network. Here, only eight variables from each of three leads were used, but the number of weights was as high as the network of this size could be trained by using a database of some 1,500 ECGS, as in this study, but much larger networks would probably not be sufficiently trained. In Contrast to it, the electrocardiographer makes his decision based on much more data. Hence, it is not surprising that the electrocardiographer outperforms the neural networks in few ECGS with minor configurational deviations, such as notches in the QRS complex. The neural networks may have been Volum:140 c 3 Figure 2. An example for input-output relationship that the networks in this study were only trained to diagnose anterior myocardial infarction. So, some ECGS with deep inverted T waves but normal QRS configuration, are likely to be classified as showing anterior myocardial infarction. This is also an another reason for misclassification. When all precordial leads are taken into account, left ventricular hypertrophy with strain is a probable diagnosis. Anyways, a neural network could only earn this pattern when number of examples is sufficient of left ventricular hypertrophy were added to the All rights Reserved 1571

5 VII. CLINICAL IMPLICATIONS:- The advantage of ANN overrule-based criteria is the enhanced diagnostic performance. Another one is the ability to easily adjust the network outputs in different clinical situations. Neural network outputs can be regarded as Bayesian aposteriori probabilities when the apriori probabilities of the classes in the training database are the same as the apriori probabilities in the test situation. An experienced electrocardiographer takes into account the clinical situation in which an ECG is recorded and adjusts the interpretation accordingly. The disadvantage in artificial neural networks is the lack of reasons for a certain diagnosis. These criteria are usually very complex because they are studied rarely in clinical practice and may not be easy for many ECG readers to understand. Nevertheless, they are well accepted by millions of users. VIII. WHERE ARE NEURAL NETWSORKS BEING USED? They are used in Signal processing: suppress line noise, with adaptive echo canceling, blind source separation Control: backing up a truck: cab position, rear position, and match with the dock get converted to steering instructions. Manufacturing plants for controlling automated machines. Siemens successfully uses neural networks for process automation in basic industries, e.g., in rolling mill control more than 100 neural networks do their job, 24 hours a day Robotics - navigation, vision recognition Pattern recognition, i.e. recognizing handwritten characters, e.g. the current version of Apple's Newton uses a neural net Medicine, storing medical records based on case information Speech production: reading text aloud (NETtalk) Vision: face recognition, edge detection, visual search engines Business, rules for mortgage decisions are extracted from past decisions made by experienced evaluators, resulting in a network that has a high level of agreement with human experts. Financial Applications: time series analysis, stock market prediction Data Compression: speech signal, image, e.g. faces Game Playing: chess, go,... IX. a)advantages: ADVANTAGES, LIMITATIONS AND APPLICATIONS OF NEURAL NETWORKS:- They can be applied to any problems, as long as there is some data. They can also be applied to problems, for which analytical methods do not exist Neural networks are used to design non-linear dependencies. If there is a pattern, neural networks will quickly work it out, even if the data is noisy. It Always gives some answer even when the input information is not complete. Networks are easy to maintain. b)limitations:- If there is no or very little data available, any data-driven models cannot be All rights Reserved 1572

6 There are many free parameters, such as the number of hidden nodes, the learning rate, minimal error, which may greatly influence the final result. Not good for arithmetics and precise calculations. Neural networks do not provide explanations. If there are many nodes, then there are too many weights that are difficult to interprete (unlike the slopes in linear models, which can be seen as correlations). In some tasks, explanations are crucial (e.g. air traffic control, medical diagnosis). X. CONCLUSION: - The Artificial Intelligence updating and extending its applications to different fields is increasing effectively. Since the field of medicine plays a vital role in the life saving activities on human, it gets more attention in both the fields. To derive final conclusion about the patient, it is mandatory to look back his past clinical history. Keeping history in the memory of a doctor is very difficult one. So the artificial Intelligence take over this function and saving the data for a long period and also reproduce it very quickly whenever it is needed. It also helps the doctors to take right decision based on the Artificial Intelligence prediction on a particular matter without any geographical barrier. There are many intelligent systems having been introduced depending upon today s needs. The present work analyzed the advantages and disadvantages and emphasized the need for future work which has to be towards more perfection and accuracy. XI. REFERENCES:- 1. Adel W.Sadek, Artificial Intelligence Applications in transportation, University of Vermont. 2. Baxt WG. Use of an artificial neural network for data analysis in clinical decision-making: the diagnosis of acute coronary occlusion. Neural Comput 1990; 2: Buchanan, Bruce G. A (very) brief history of Artificial Intelligence. AI Magazine 26(4): Winter 2005, Cohen, Jonathan. Human robots in myth and science. NY: A.S.Barnes, Feigenbaum, E.A. & Feldman, J. (eds.) Computers and Thought. NY: McGraw-Hill, Gardner, Martin. Logic Machines and Diagrams NY: McGraw-Hill, Gunn AA. The diagnosis of acute abdominal pain with computer analysis. J R Coll Surg Edinb 1976; 21: Ledley RS, Lusted LB. Reasoning foundations of medical diagnosis. Science 1959; 130: McCorduck, Pamela. Machines who Think. San Francisco: W.H. Freeman, Minsky, M. L., and S. Papert. Perceptrons: An Introduction to Computational Geometry (First edition). 11. MIT Press, Cambridge, Massachusetts, Pranav Deshpande works with Startup Bridge at Stanford, an annual conference held in December at Stanford where top technology innovators from India and Silicon Valley build strategic partnerships to innovate for the world. 13. Russel, S. J., and P. Norvig. Artificial Intelligence: A Modern Approach. Prentice-Hall, Upper Saddle River, New Jersey, Stamey TA, Barnhill SD, Zang Z. Effectiveness of ProstAsureTM in detecting prostate cancer (PCa) and benign prostatic hyperplasia (BPH) in men age 50 and older. J Urol 1996; 155: 436A. 15. Zadeh, L. A. Fuzzy Sets. Information and Control. Vol. 8, 1965, pp. 338 All rights Reserved 1573