Clustering and screening for breast cancer on thermal images using a combinati on of SOM and MLP

Breast screening is a valuable method of decreasing chances of death from breast cancer, which is the most frequent cancer in women. Several methods are currently used to screen for breast cancer. The proposed method in this paper applies artificial intelligence to detect and screen breast cancer using thermal images, in order to minimise the possibility of the physician’s diagnosis errors. First, data on a thermal image taken from a patient are clustered by using selforganising neural networks for this purpose. Then, suspicious areas from the image are separated. The results of this step are used in an algorithm that is similar to the basic algorithm (self-organising map algorithm of primary suggested), but it has different specifications to extract diagnostic features for screening. Finally, these specificities are fed into the multilayer perceptron neural network to complete the screening process. The images considered for testing includes two 200 case bases and one 50 case bases. In the former 15 cases and in the latter 2 cases were diagnosed with cancer by mammography. The results of the first and second bases show sensitivities of 88% and 100% (accuracy ¼ 98.5%), respectively.

breast cancer screening; self organised map; multi-layer perceptron neural networks