Cystoscopic Image Classification Based on Combining MLP and GA

Abstract

In the past three decades, the use of smart methods in medical diagnostic systems has attracted<br /> the attention of many researchers. However, no smart activity has been provided in the field of<br /> medical image processing for diagnosis of bladder cancer through cystoscopy images despite the high<br /> prevalence in the world. In this paper, a multilayer neural network was applied to classify blad-<br /> der cystoscopy images. One of the most important issues in training phase of neural networks is<br /> determining the learning rate. Because selecting too small or large learning rate leads to slow con-<br /> vergence, volatility and divergence, respectively. Therefore, an algorithm is required to dynamically<br /> change the convergence rate. In this respect, an adaptive method was presented for determining the<br /> learning rate so that the multilayer neural network could be improved. In this method, the learning<br /> rate is determined using a coefficient based on the difference between the accuracy of training and<br /> validation according to the output error. In addition, the rate of changes is updated according to the<br /> level of weight changes and output error. Another challenge in neural networks is determining the<br /> initial weights. In cystoscopy images, randomized initial weights should not be used due to a small<br /> number of images collected. Therefore, the genetic algorithm (GA) is applied to determine the initial<br /> weight. The proposed method was evaluated on 540 bladder cystoscopy images in three classes of <br /> blood in urine, benign and malignant masses. Based on the simulated results, the proposed method<br /> achieved a 7% decrease in error and increased the convergence speed of the proposed method in the<br /> classification of cystoscopy images, compared to the other competing methods.