Diagnosis of Breast Cancer using SOM Neural Network

Abstract A neural network system is designed and optimize to analyze the quantitative data from the impedance spectrum from where the breast tissue features are computed. These data was used to predict and classify the breast cancer Car (carcinoma), fad (fibro-adenoma+ mastopathy + glandular), Con (connective), Adi (adipose). The performance of an artificial neural network (ANN) is verified with nine quantitative parameters computed from Impedance measurements. The Self Organizing Feature Map (SOM) network is trained using the different data partitioning methods and tested its performance on seen and unseen data in terms of classification accuracy, MSE and correlation coefficient. The network is yielded better classification accuracy (93.75%) with testing and cross validation MSE of 0.00044 and 0.00025 respectively.