Microwave imaging method employing wavelet transform and neural networks for breast cancer detection

The combined use of wavelet transform and neural networks to detect and diagnose early breast cancer is investigated. The utilized neural network is of the feed-forward back-propagation type. The proposed algorithm is tested on a three-dimensional heterogeneous breast model. Spherical tumors of radii 1mm, 3mm, and 5mm are assumed at different locations in the breast model. An ultrawideband pulse is transmitted towards the breast model and twenty-four probes are located around the breast to capture the scattered signals. The collected signals are then analyzed using the wavelet transform combined with neural networks to get useful information concerning the presence of the tumor and its size if it does exist. The detection capability of the presented method depends on the extracted features from the scattered signals. The features adopted here are cross-correlation between reflected and transmitted signal, standard deviation, mean, and energy of each sub-band of the utilized wavelet. To get those features, five levels of wavelet analysis are used. The obtained results from using the proposed method are promising with 100% success in the tumor detection. The rates of correct recognition of tumor size are 100%, 76.56%, and 65.52% for tumor radii of 5 mm, 3 mm, and 1 mm, respectively.