Classification of Mammograms Using a Modular Neural Network
T. Cooley, E. Micheli-Tzanakou · Journal of Intelligent Systems · 1998
The number of deaths linked to breast cancer has risen to 44,000 women annually.Early detection and treatment is the key to long term survival.Mammography remains the number one method for detection of breast cancer.Studies have shown that radiologists may differ in their interpretation of mammograms.A system is described which uses a digitized image of a mammogram and classifies it as normal or abnormal without the aid of an expert.This computer system processes the given image using multiresolution wavelet analysis and then extracts features using invariant moments.The extracted features are then used as inputs to a modular neural network that classifies the mammogram.Training results of 93.9 percent and validation results of 100 percent were achieved on a small data set.