Mametsaliyev R.R.
NEURAL NETWORKS IN THE RECOGNITION AND DIAGNOSIS OF MEDICAL DISEASES. *
Аннотация:
this article discusses the features of the use of neural networks in the recognition of medical diseases and their influence on the processes of studying diseases. A cross and comparative analysis of the influence of the choice of the direction of technology development in medicine was carried out. Recommendations are given for the implementation of developments in the study of the theory of neural networks
Ключевые слова:
analysis, method, research, medicine, neural networks
УДК 616.67
Mametsaliyev R.R.
Lecturer,
Engineering and Technology University of Turkmenistan named after Oguzhan
(Turkmenistan, Ashgabat)
NEURAL NETWORKS IN THE RECOGNITION
AND DIAGNOSIS OF MEDICAL DISEASES
Abstract: this article discusses the features of the use of neural networks in the recognition of medical diseases and their influence on the processes of studying diseases. A cross and comparative analysis of the influence of the choice of the direction of technology development in medicine was carried out. Recommendations are given for the implementation of developments in the study of the theory of neural networks.
Keywords: analysis, method, research, medicine, neural networks.
The rapid improvement of medicine, the unreadable flow of incoming information about the etiology and pathogenesis of diseases dictate the rules for finding new methods for processing the results obtained. Given this, you need to be able to find the necessary information and form the right decisions that will affect the course and outcome of the disease. This is largely facilitated by the introduction of computer technologies, the creation of software allows us to analyze the tasks of verification, diagnosis and prediction of various diseases, as well as the search for optimal methods of treatment. One of the promising areas of medical neuroinformatics is neural networks (NNs). The concept of a neural network was defined in 1943 when W McCullogue and W Pitts organized the first mathematical model of a neural network. A practical model using a computer was created in 1957 by F. Rosenblatt. Since then, the neural network has been actively used in various areas of society and science [1-4].
The issue of predicting and verifying the diagnosis is always relevant when a doctor deals with chronic recurrent diseases or a certain staging of the course of inflammation is observed. Despite the fact that there are a huge number of forecasting methods in the world, there is no complete automated mechanism that allows to carry out this task in a short time. Moreover, the individual characteristics of the patient's body, age, dynamics of the disease and other indicators make it impossible to use these techniques in practice. It is also important to take into account the volume of statistical material (sample), loss of information, data omissions and statistical errors. As a result, the result of the forecast may be unreliable. As an example, it is relevant to cite acute appendicitis, which occurs as a catarrhal inflammation turning into phlegmonous, gangrenous forms of inflammation. Similarly, with regard to acute cholecystitis (the catarrhal form turns into a phlegmonous, then into a gangrenous form). In clinical practice, it is quite difficult to determine the transition of one form of inflammation to another.
In this paper, we analyzed the work and processed data from a retrospective study of case histories of patients who received inpatient treatment in a number of private clinics; to solve the problem, patient data were anonymized; therapeutic profile pathologies were predicted using modern statistical software packages and the MATLAB modeling environment. A recurrent (dynamic) NN was used, which implements the dependence of the form. Figure 1 shows the architecture of a recurrent NS, where Z–1 is the time delay element for one cycle Δt; N is the number of delay elements (the size of the "time window"); yi (t+1) is the predicted (one step ahead) value of the parameter yi (t). Let us consider in more detail the description of the method for predicting the dynamics of the development of pathologies, built on the basis of extrapolating functions yi (t) as a function of time: yi (t) = f(t), (4) where t is the current time. The implementation of this approach in the neural network basis is carried out as follows: - a time interval (observation interval) is allocated, which is a training sample for the neural network (t is the input of the neural network; parameters y1, y2, ..., yn of the patient are its outputs); - the forecast step is set - Tprogn. taking into account the requirements for the forecast (short-term, medium-term, long-term forecast); - after the learning process of the NN on the observation interval, the predicted values yi (t+T prog) are calculated, for this the time value (t+T prog) is fed to the input of the NN; - then the forecasting process is repeated in real time.
The results of experimental studies of the NN reflect the minimum learning error of the NN over an observation interval containing 12 samples, provided that the number of neurons in the hidden layer is six. So, in the case under consideration, the optimal structure of the neural network has the form: 27–6–1 (i.e., 27 neurons are used at the input, 6 in the hidden layer, 1 neuron in the output layer). As a "time window" that forms the predicted series, it is advisable to set 17 elements.
The creation of an operational algorithm of work is the main task in the implementation of the work of a neural network, initially it is necessary to determine the architecture of the neural network. Medical parameters have a large amount of input data, and, consequently, a huge sample and a large number of parameters, so it is most advisable to choose a 3-layer recurrent feed-forward network.
Picture 1.
- where N is the number of previous periods included in the moving average; y{ is the actual value at time y is the predicted value at time (t+1).
It is advisable to choose a 3-layer recurrent feedforward network as the architecture of the neural network. In the case under consideration, the optimal structure of the neural network has the form: 27-6-1 (i.e. 27 neurons are used at the input, 6 in the hidden layer, 1 neuron in the output layer).
When evaluating the effectiveness of neural network prediction of the dynamics of the development of diseases of a therapeutic profile, a comparative analysis was carried out with the method of group consideration of the argument.
The use of the neural network methods proposed in the work for predicting the dynamics of the development of pathologies makes it possible to effectively solve the problems of forecasting. In view of the foregoing, it can be concluded that the use of the neural network method presented in the paper for predicting the dynamics of the development of pathologies makes it possible to effectively solve forecasting problems.
REFERENCES:
Номер журнала Вестник науки №6 (63) том 1
Ссылка для цитирования:
Mametsaliyev R.R. NEURAL NETWORKS IN THE RECOGNITION AND DIAGNOSIS OF MEDICAL DISEASES. // Вестник науки №6 (63) том 1. С. 863 - 866. 2023 г. ISSN 2712-8849 // Электронный ресурс: https://www.вестник-науки.рф/article/8689 (дата обращения: 19.05.2024 г.)
Вестник науки СМИ ЭЛ № ФС 77 - 84401 © 2023. 16+