Investigation of Genetic Algorithms for Computer-Aided Diagnosis

摘要

Computer-aided diagnosis has the potential of substantially increasing diagnostic accuracy in mammography. Using a computer to double-check a radiologist's findings is becoming more popular and more important as the public learns that the best defense against breast cancer is early detection. The University of Chicago is currently developing computerized schemes to detect cancers in digital mammograms. We use a pattern classification system known as an artificial neural network (ANN) to classify certain regions of the digital mammograms as cancerous or non-cancerous. ANNs are trained pattern classification devices that take, as inputs, features extracted from regions in the mammograms and output the classification. Currently, there are a total of 42 features extracted from the various regions in each mammogram. A subset of those 42 features must be chosen as inputs for the ANN. The goal of this research was to investigate methods of feature selection and pattern classification in order to improve upon the overall performance of CAD systems.