Hybrid Data Ananlysis Methods and Artificial Neural Network Design in Breast Cancer Diagnosis: IDEST Experience
Vitoantonio Bevilacqua, Giuseppe Mastronardi, Filippo Menolascina · 2006
This paper presents a method for breast cancer diagnosis using a system based on an artificial neural network (ANN) trained using a particular version of the back propagation (BP) algorithm. The Wisconsin Breast Cancer Database (WBCD) was used in order to train and validate the ANN; WBCD is composed by 699 cases monitored by Doc. William H. Wolberg in the first '90s. The development of this system required an articulate phase of data analysis and preprocessing: various statistical tools like principal component analysis (PCA) and principal factor analysis (PFA), were used in order to find parameters more strictly correlated to the malignant/benignant nature of the cancer. Non linear data analysis techniques were employed to gain more knowledge about the internal structure of the database. A genetic algorithm was then set up to find the best topology of ANN. The analysis of results obtained by IDEST followed training and validation of the ANN