The application of pattern recognition techniques to radiology has the potential to detect cancer earlier and save lives, and consequently much research has been devoted to this problem. This worked tackled a subset of the problem, investigating a novel method of classifying mammograms using an evolutionary approach known as Cartesian Genetic Programming (CGP). Microcalcifications, one of two major indicators of cancer on mammograms, were used for the classification. A large software framework was written in order to investigate this, which allows the viewing of images, manual segmentation of lesions and then automatic classification. Two classification approaches were pursued, the first classifying on texture features and the second, a new approach, classifying by using the lesion's raw pixel array. Early results using the system showed some potential. It was found that during training, networks could obtain correct classification rates of between 80 and 100%. The best results were approaching those in the contemporary literature and suggest the technique warrants further investigation.
Evolutionary Algorithms in the Classification of Mammograms
2007-04-01
9413158 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Microcalcification classification in mammograms using multiwavelet features [3813-87]
British Library Conference Proceedings | 1999
|Autoradiographic enhancement of mammograms
NTRS | 1979
|Evaluating Classification Strategies for Detection of Circumscribed Masses in Digital Mammograms
British Library Conference Proceedings | 1999
|Segmentation of mammograms in topographic approach
British Library Conference Proceedings | 2003
|Circumscribed Mass Detection in Digital Mammograms
IEEE | 2006
|