Abstract: The economic and social values of breast cancer diagnosis are very high. This study explores the predictive abilities of Fuzzy ARTMAP neural networks for breast cancer diagnosis. The data used is a combination of 39 mammographic, sonographic, and other descriptors, which is novel for the field. By using feature selection techniques we propose a subset of 21 descriptors that outperform the full feature set and outperforms the prediction model based on the most popular MLP neural networks. We also explored the model performance by ROC analysis and used metrics, such as max accuracy, area under the ROC curve, and area under the convex hull. Due to lack of specificity, many diagnosis tools entail unnecessary surgical biopsies, which motivated us to explore the clinically relevant metrics partial area under the ROC curve where sensitivity is above 90% and specificity at 98% sensitivity. In conclusion we find that the Fuzzy ARTMAP neural network is a promising prediction tool for breast cancer diagnosis. To the best of our knowledge, the Fuzzy ARTMAP neural networks have not been studied in that area until now.

Keywords: data mining, neural networks, Fuzzy ARTMAP, heterogeneous data; breast cancer diagnosis, computer aided diagnosis

ACM Classification Keywords: I.5.1- Computing Methodologies - Pattern Recognition – Models - Neural Nets

Link:

FUZZY ARTMAP NEURAL NETWORKS FOR COMPUTER AIDED DIAGNOSIS

Anatoli Nachev

http://foibg.com/ibs_isc/ibs-20/ibs-20-p01.pdf