Comparison of back propagation neural network with hybrid back propagation-wavelet network for breast cancer diagnosis, based on statistical features extracted from thermographic images of women's breasts.

Abstract

In order to diagnose breast cancer, methods such as mammography, MRI, thermal mammography and detection with a simple breast health test device (Brest Angel) are used. Different image processing methods are among the effective methods for detecting different types of tumors in women's breasts. In this article, two types of artificial neural networks are used. 5 statistical features extracted from thermographic images of women's breasts were used to diagnose cancer in neural networks. In this article, back propagation neural network (network 1) is used with Lunberg-Markudat training method and its results are compared with hybrid back propagation-wavelet network (network 2) to investigate the condition of women's breasts. The outputs of the two neural networks used in the article have 2 nodes, which indicate whether the person in question has breast cancer or not with the information given to the neural networks. In network (1), correlation coefficient (R=0.9831) and root mean square error (RMSE=0.5538) were obtained as the best function for network training. In contrast to the network correlation coefficient (2), R=0.9945 and root mean square error (RMSE=0.4665) was obtained. The training time of neural network 1 was 45.51 seconds and network 2 was 33.68 seconds. The results of the designed wavelet-back propagation hybrid neural network show that the proposed network is effective in detecting breast cancer with 99.5% accuracy and is able to detect the health status of women's breasts.

Keywords

• breast cancer
• artificial neural network
• hybrid-back propagation neural network
• statistical feature
• breast thermography images