Application of artificial neural networks in medicine Elda Xhumari Petrika Manika Department of Informatics Department of Informatics Universtity of Tirana, TIRANA 1001 Universtity of Tirana, TIRANA 1001 elda.xhumari@fshn.edu.al petrika.manika@fshn.edu.al Research Center in Milan. The program enabled non- Abstract invasive cardiodiagnostic, based on the spectrum recognition of tahograph. Tahograph histogram The use of neural networks in medicine, represents the intervals between heart beats and its normally is linked to disease diagnostics activity spectrum reflects the balance of sympathetic and systems. However, neural networks are not parasympathetic nervous system, which differs only able to recognize examples, but maintain very important information. For this reason, specifically for various diseases. Now we can conclude one of the main areas of application of neural that neural networks have become an instrument of networks is the interpretation of medical data. diagnosis of heart disease - in UK, for example, used in In this article we will discuss the application of four hospitals for the prevention of myocardial infarct. neural networks in medicine with a concrete Neural networks can also be used to forecast the action example - a diagnosis of diabetes disease in its of various healing treatments. Successfully applied in early stages. chemistry for predicting molecules properties of different interactions. 1. Introduction In this article we will discuss the application of neural Neural networks are nonlinear systems, which make it networks for diagnosing diabetes disease in its early possible to classify the data better than linear methods. stages. Nowadays, diabetes is considered one of the most A distinctive feature of neural networks is that they are prevalent diseases in the world. According to the World not programmed - do not use any rule for diagnosis, but Health Organization, around 30 million people of are trained to do so in the examples. Regarding this, various ages and breeds suffer from various forms of neural networks aren’t similar to expert systems, the diabetes1. Diabetes is not a consequence of any development of which in the 70s was held after a particular organ pathology, but in general displayed by temporary "victory" of artificial intelligence approach the metabolic disorder. His symptoms occur in organs or against memory modeling, recognition and synthesis organ systems, which are more vulnerable to this models, which was based on the study of neural process. Clinical signs of diabetes depend on the type of organization of human brain. disease, gender, age, level of insulin, arterial pressure One of the most prominent and developed, the effect of and many other factors. which was based on the knowledge obtained from the Neural network technologies are designed to solve many experts, was Streptomycin system. This system was difficult tasks, starting from formulation, among which developed in Stanford in the early '70s for the diagnosis many medical problems. This is related to the fact that of septic shock. Half of the patients died, and doctors to the researchers are often given a large number of could reveal sepsis only in 50% of cases. Streptomycin factual materials, for which there is no mathematical was apparently a real triumph of expert systems model. Artificial neural networks models have shown technology - because it enabled the detection of sepsis- good results for diagnose nerve disorders [Hul1], it in 100% of cases. As a diagnostic program sample Parkinson's disease [Gil2] and Huntington [Sin3] served cardiodignostic package, developed by the RES diagnosis. Multilayer perceptron models are used for company together with Informatica Cardiology predicting the risk of occurrence of osteoporosis [Bas4]. For Hepatitis B diagnostic is used generalized 1 10 facts about diabetes – www.who.int/features/factfiles/diabetes/en. mathematical regression [Mah5]. One of the most useful 1 𝐹= , 1+exp⁡(−𝛼𝑌) tools for solving the tasks above are artificial neural networks - a powerful simulation method of phenomena where α is a coefficient chosen experimentally. and processes at the same time very flexible. Modern neural networks are a combination of devices and programs designed for models and specialized equipment for solving a wide range of diagnostics tasks by applying a set of algorithms theory of object recognition. The distinguishing feature of neural networks is the ability to be trained in a specific field. Applied to medical topics, experimental data represent a set of input parameters of the object (in our case health parameters). Training network neural is an interactive process, at the entrance of which net finds secret nonlinear relationships between input parameters and final diagnosis, also the optimal combination of weight Figure 1: Sigmoidal activation function coefficients of neurons, joining the neighbour layers for which the error of definig sample class goes to minimum Multi-layer perceptron has a high level of connections, [Yak6]. As one of the advantages of neural networks we which are realized through synapses. The change of level should consider their relative simplicity, non-linearity, of network connections requires changing of synaptic working with incomplete information, the ability to train connections or coefficients of their weights. Combining on concrete examples. During the training process, as all the features together with the ability to train their initial input we will give parameters together with the experience, provides the computing power of multi- diagnosis, characterizing these parameters. To train a layer perceptron. neural network we must have a sufficient number of Artificial neural network consists of 3 layers: input, examples to adapt the system to a certain level of secret and output layer. credibility. If the examples are dealing with various Input layer has 12 neurons, while output has only 2. diagnostics, then artificial neural network trained in this (figure 2). way allows the diagnosis of each new case, presented by a group of indicators, similar to indicators used to train the others. The apparent advantage of neural model is that the creation is not necessary to present the complex connections to describe the phenomenon that is diagnosed. During the application of neural networks for solving various practical tasks we may face a serie of difficulties. One of the key problems is the application of these technologies and the large degree of difficulty of the network design (not known whether it will be right enough to diagnose disease). This difficulty can complicate network architecture. Simple nets with only Input Secret Output layer layer layer one layer (single-layer perceptron) are able to solve only divisible, linear tasks [Gal7]. This limitation can be Figure 2: Architecture of neural network overcome using multi-layer networks (multilayer perceptron). Table 1. Input-layer parameters 2. Experiment Parameter Type of data, unit For this article we used a multi-layer model and backpropagation algorithm for training. As activation Age Number (years) function - sigmoidal function (Figure 1). Physical loads Logic (Yes/No) Sex Logic (Female/Male) Number of pregnancies Number Diabetes to relatives Logic (Yes/No) algorithm steps are repeated until we receive the desired Body mass index Number (kg/m2) error. Skin thickness Number (mm) Colesterin level Number (mg/dl) 3. Conclusions Diastolic pressure Number (mm Hg) Trained and tested database consisted of data on 486 Insulin Number (µUnit/ml) patients, among whom 243 were given the diagnosis Stress Logic (Yes/No) "diabetes", while the rest were healthy. Glucose level Number (mg/dl) Were trained over 240 examples and neural network was tested in 146 samples. The reliability of the model For the design of neural network we used the package of amounted to 89.5%. Theoretical difficulty, workload Matlab Neural Network Toolbox R2012a. The package and time lost for network modeling were compensated represents a set of functions and data structure, which by the simplicity of model using. If the task of solving describes the functions of activation, training the problem and optimal training are carried out only by algorithms, setting of synaptic connections, etc. specialists, then the practical application of the solution requires only basic knowledge of computer use. The difficulty of interpreting the train system for users deemed unnecessary simple, given that their most important is not the way of functioning of the neural network, but the result information, accuracy and operation speed. References Figure 1: Functional scheme of backpropagation algorithm [Hul1] Hulshoff Pol H1, Bullmore E. Neural networks The algorithm used (backpropagation) is assumed to in psychiatry - www.ncbi.nlm.nih.gov. estimate the error for output layer, as well as for each of [Gil2] Gil D., Johnsson M. Diagnosing Parkinson by trained network neuron, also neurons correction weights using artificial neural networks and support vector in accordance with current values. At the first step of the machines. Global Journal of Computer Science and algorithm we initialized weights of connections of Technology, 2009, №9(4). pp.63-71. intermediate neurons with a negligible value (from 0 to [Sin3] Singh M., Singh M., Singh P. Artificial Neural 1). After initialization of the weights in the training Network based classification of Neuro-Degenerative process of neural network met steps: diseases using Gait features. International Journal of - direct signal delivery; Information Technology and Knowledge Management, - error calculation for last layer neurons; 2013, Vol. 7, №1, pp. 27-30. - error delivery in the opposite direction (from [Bas4] Basit A., Sarim M., Raffat K., etc. Artificial output to input). Neural Network: A Tool for Diagnosing Osteoporosis. Direct signal delivery is carried by layer, starting from Research Journal of Recent Sciences, 2014, Vol. 3(2), the entry layer, calculating the amount of the input pp.87-91. signals for each neuron and with the help of activation [Mah5] Singh M., Singh M., Singh P. Artificial Neural function is generated a response of the neuron, which is Network based classification of Neuro-Degenerative distributed in the next layer heeding the weight of diseases using Gait features. International Journal of connection between neurons. The next stage of training Information Technology and Knowledge Management, - error calculation of neural network as the difference 2013, Vol. 7, №1, pp. 27-30. between actual and desired. [Yak6] Yakh''yaeva G.E. Nechetkie mnozhestva i Error values obtained are distributed from the last output neironnye seti. M.: Internet-Universitet layer to the first layer of neural network. For this we informatsionnykh tekhnologii: BINOM. Laboratoriya calculate correction values for neuron weights znanii, 2006. – 316 s. depending on the current value of the connection [Gal7] Galushkin A. Neironnye seti. Osnovy teorii. between neurons, speed training and error for the Goryachaya Liniya – Telekom, 2012. - 496 s. corresponding neuron. After fulfilling this stage, the