Comparison of the Particle Swarm Optimization with the Genetic Algorithms as a Training for Multilayer Perceptron Technique to Diagnose Thyroid Functional Disease

Abstract

Background: Thyroid is a gland that controls body key functions. Since thyroid hormones are responsible for controlling metabolism, the thyroid glands are one of the major organs in the body. Disruption of the glands causes thyroid disease, which is one of the most common endocrine diseases worldwide. Early diagnosis of thyroid disease is difficult since early symptoms are easily confused with those of other illnesses. In recent years, the employment of computer techniques to utilize data mining and intelligent algorithms accelerates the early diagnosis of the disease. Objectives: The current study aimed at evaluating the role of the new automatic method according to a multilayer perceptron (MLP) in the diagnosis of thyroid disease. Methods: The study aimed at comparing the particle swarm optimization (PSO) with the genetic algorithms (GA) as training for MLP technique used to diagnose thyroid functional disease. The data were collected for three classes: 150 cases of euthyroidism, 30 cases of hypothyroidism, and 35 cases of hyperthyroidism. MLP was used to elucidate the pattern in the data and species responsible for separating the classes. Furthermore, improved PSO and GAs were used to train the system, and the sensitivity and specificity of the model were studied in terms of accuracy, sensitivity, and specificity. All analyses were performed using MATLAB software. Results: For the proposed model, the simulations results showed that the GA algorithm had a higher performance than the PSO algorithm in the diagnosis of functional thyroid disease, and the means of classification accuracy, sensitivity, and specificity were 95%, 96%, and 96%, respectively. Conclusions: The results of real data indicated that the GA-MLP can be used with high diagnostic accuracy as an effective tool to clinically diagnose thyroid functional disease. The current study was a step towards prototype system development of the classification of knowledge in this area with a much lower computational cost.