برآورد هزینه‌ی درمان و طول دوره‌ی بستری شدن با استفاده از رویکرد شبکه‌ی‌ عصبی

Abstract

Background: Using neural networks and genetic algorithms in evaluating health-related variables has increased recently. Employing intelligent tools for diagnosis and treatment of diseases can reduce medical errors and human and financial losses. In this paper, medical applications of neural networks have been studied in order to help both medical and artificial intelligence researchers. Methods: We used an existing sample in SPSS (patient_los.sav). The sample consisted of patients who received treatment for heart disease. Multilayer perceptron (MLP) was employed to build a neural network to predict the cost and length of treatment. Duration of hospitalization and treatment cost were considered as dependent variables. Other variables were entered into the model as agents or factors. Results: Neural networks can evaluate the outcomes of patients who have or have not undergone surgery. Separate networks can then be used to predict treatment and hospitalization costs and duration provided that the patients who had surgery had been identified. Conclusion: Neural networks designed in this paper can well forecast the usual outcomes of patients. Multilayer neural networks can precisely identify patients who would die after surgery. Non-linear properties of neural networks can help in modeling and forecasting. Keywords: Neural Networks (Computer); Diagnosis; Learning.