A domain of interest for data mining applications is the study of biomedical data which, in combination with the field of image processing, provide thorough analysis in order to discover hidden patterns or behavior. Towards this direction, the present paper deals with the detection of breast cancer within digital mammography images. Identification of breast cancer poses several challenges to traditional data mining applications, particularly due to the high dimensionality and class imbalance of training data. In the current approach, genetic algorithms are utilized in an attempt to reduce the feature set to the informative ones and class imbalance issues were also dealt by incorporating a hybrid boosting and genetic sub-sampling approach. As regards to the feature extraction approach, the idea of trainable segmentation is borrowed, using Decision Trees as the base learner. Results show that the best precision and recall rates are achieved by using a combination of Adaboost and k-Nearest Neighbor.