Identifying Input Mapping Features

To implement a knowledge-based contrast enhancement component to learn the target enhancement for a given mammogram, a supervised learning paradigm is employed. By utilizing pattern recognition tools, a classifier can learn from a set of example mammograms the target contrast enhancement. For an unseen testing mammogram, the trained classifier will accurately predict the actual enhancement that maximizes segmentation performance.

During training the classifier learns a mapping between a characteristic of an example training mammogram and the target enhancement method. To facilitate this mapping, features are extracted to characterize the training and testing mammograms. Two different approaches to feature extraction are described: (1) feature extraction from a ROI and (2) feature extraction from a breast profile.

0 0

Post a comment